Compare commits

...

14 Commits

Author SHA1 Message Date
jkunz 2b9a17c882 v3.7.0
Default (tags) / security (push) Failing after 1s
Default (tags) / test (push) Failing after 1s
Default (tags) / release (push) Has been skipped
Default (tags) / metadata (push) Has been skipped
2026-05-18 08:53:59 +00:00
jkunz 9957485281 feat(mcp): add MCP tool integration via dedicated subpath export 2026-05-18 08:53:57 +00:00
jkunz cf45add905 v3.6.0
Default (tags) / security (push) Failing after 1s
Default (tags) / test (push) Failing after 1s
Default (tags) / release (push) Has been skipped
Default (tags) / metadata (push) Has been skipped
2026-05-15 08:03:17 +00:00
jkunz 7b4184e88d feat(readme): document provider prompt caching behavior and SmartAI cache integration 2026-05-15 08:03:15 +00:00
jkunz 01df877480 v3.5.0
Default (tags) / security (push) Failing after 1s
Default (tags) / test (push) Failing after 1s
Default (tags) / release (push) Has been skipped
Default (tags) / metadata (push) Has been skipped
2026-05-15 05:38:50 +00:00
jkunz f138495edf feat(tools): add reusable execution contexts for shell, filesystem, and browser tools 2026-05-15 05:38:35 +00:00
jkunz f183bf19ac v3.4.0
Default (tags) / security (push) Failing after 1s
Default (tags) / test (push) Failing after 1s
Default (tags) / release (push) Has been skipped
Default (tags) / metadata (push) Has been skipped
2026-05-14 22:44:10 +00:00
jkunz 6fb2b3a61f feat(agent): add streamed reasoning summary callbacks to runAgent 2026-05-14 22:44:08 +00:00
jkunz ca56f4c4e8 v3.3.0
Default (tags) / security (push) Failing after 1s
Default (tags) / test (push) Failing after 1s
Default (tags) / release (push) Has been skipped
Default (tags) / metadata (push) Has been skipped
2026-05-14 16:50:16 +00:00
jkunz 5ceeddd8bb feat(deps): upgrade @push.rocks/smartai to ^4.0.0 2026-05-14 16:50:08 +00:00
jkunz d7edb981e7 v3.2.0
Default (tags) / security (push) Failing after 1s
Default (tags) / test (push) Failing after 1s
Default (tags) / release (push) Has been skipped
Default (tags) / metadata (push) Has been skipped
2026-05-14 11:34:11 +00:00
jkunz e6346be884 feat(agent): add prompt caching options and cache token usage reporting 2026-05-14 11:34:04 +00:00
jkunz 7be67543bf v3.1.1
Default (tags) / security (push) Failing after 1s
Default (tags) / test (push) Failing after 1s
Default (tags) / release (push) Has been skipped
Default (tags) / metadata (push) Has been skipped
2026-05-11 11:11:43 +00:00
jkunz 28b9b215f3 fix(smartconfig): update release configuration to schema version 2 with npm target settings 2026-05-11 11:11:40 +00:00
20 changed files with 2134 additions and 201 deletions
+12 -7
View File
@@ -11,12 +11,17 @@
"projectDomain": "push.rocks"
},
"release": {
"registries": [
"https://verdaccio.lossless.digital",
"https://registry.npmjs.org"
],
"accessLevel": "public"
}
"targets": {
"npm": {
"registries": [
"https://verdaccio.lossless.digital",
"https://registry.npmjs.org"
],
"accessLevel": "public"
}
}
},
"schemaVersion": 2
},
"@git.zone/tsdoc": {
"legal": "\n## License and Legal Information\n\nThis repository contains open-source code that is licensed under the MIT License. A copy of the MIT License can be found in the [license](license) file within this repository. \n\n**Please note:** The MIT License does not grant permission to use the trade names, trademarks, service marks, or product names of the project, except as required for reasonable and customary use in describing the origin of the work and reproducing the content of the NOTICE file.\n\n### Trademarks\n\nThis project is owned and maintained by Task Venture Capital GmbH. The names and logos associated with Task Venture Capital GmbH and any related products or services are trademarks of Task Venture Capital GmbH and are not included within the scope of the MIT license granted herein. Use of these trademarks must comply with Task Venture Capital GmbH's Trademark Guidelines, and any usage must be approved in writing by Task Venture Capital GmbH.\n\n### Company Information\n\nTask Venture Capital GmbH \nRegistered at District court Bremen HRB 35230 HB, Germany\n\nFor any legal inquiries or if you require further information, please contact us via email at hello@task.vc.\n\nBy using this repository, you acknowledge that you have read this section, agree to comply with its terms, and understand that the licensing of the code does not imply endorsement by Task Venture Capital GmbH of any derivative works.\n"
@@ -24,4 +29,4 @@
"@ship.zone/szci": {
"npmGlobalTools": []
}
}
}
+72 -1
View File
@@ -1,5 +1,76 @@
# Changelog
## Pending
## 2026-05-18 - 3.7.0
### Features
- add MCP tool integration via dedicated subpath export (mcp)
- introduces a new @push.rocks/smartagent/mcp entrypoint with MCP client, stdio, HTTP, and custom transport support
- adds MCP tool discovery, name sanitization, multi-server prefixing, result formatting, and cleanup handling
- documents MCP usage in the README and adds test coverage for tool exposure, pagination, and result formatting
## 2026-05-15 - 3.6.0
### Features
- document provider prompt caching behavior and SmartAI cache integration (readme)
- Adds provider prompt caching to the feature list in the README
- Clarifies that provider-specific cache metadata is centralized in @push.rocks/smartai
## 2026-05-15 - 3.5.0
### Features
- add reusable execution contexts for shell, filesystem, and browser tools (tools)
- introduces shared tool execution context interfaces plus a local Node.js context implementation
- adds createShellTools, createFilesystemTools, and createBrowserTools for host-provided transports with permission hooks
- re-exports the new tool factories and context types from the main package entrypoint while keeping compatibility wrappers
## 2026-05-14 - 3.4.0
### Features
- add streamed reasoning summary callbacks to runAgent (agent)
- Introduces onReasoningStart, onReasoningDelta, and onReasoningEnd callbacks in the agent options interface
- Handles reasoning-start, reasoning-delta, and reasoning-end stream chunks while accumulating reasoning text by id
- Ensures incomplete reasoning streams are finalized after the response completes
- Adds tests for reasoning summary streaming and updates the README API documentation
## 2026-05-14 - 3.3.0
### Features
- upgrade @push.rocks/smartai to ^4.0.0 (deps)
- Updates the core smartai dependency from ^2.3.0 to ^4.0.0.
- Refreshes README hints to document the new smartai version.
## 2026-05-14 - 3.2.0
### Features
- add prompt caching options and cache token usage reporting (agent)
- adds sessionId and cache run options to configure provider-specific prompt caching defaults
- applies OpenAI cache provider options and Anthropic cache breakpoints automatically, with support to disable defaults
- extends usage reporting to include cacheReadTokens and cacheWriteTokens
- exports cache-related types and helpers and updates tests and README to cover the new behavior
## 2026-05-11 - 3.1.1
### Fixes
- update release configuration to schema version 2 with npm target settings (smartconfig)
- migrates release settings from a flat registries/accessLevel structure to a nested targets.npm configuration
- adds schemaVersion 2 to align the smartconfig format with the updated release schema
## 2026-05-07 - 3.1.0 - feat(agent)
add provider options passthrough, tool call records, and completion validation retries
@@ -188,4 +259,4 @@ Bump version to 1.0.2 (patch release)
Initial commit: project scaffold and first release.
- Repository initialized with initial project structure and baseline files.
- Version set to 1.0.1.
- Version set to 1.0.1.
+9 -3
View File
@@ -1,6 +1,6 @@
{
"name": "@push.rocks/smartagent",
"version": "3.1.0",
"version": "3.7.0",
"private": false,
"description": "Agentic loop for ai-sdk (Vercel AI SDK). Wraps streamText with stopWhen for parallel multi-step tool execution. Built on @push.rocks/smartai.",
"main": "dist_ts/index.js",
@@ -18,6 +18,10 @@
"./compaction": {
"import": "./dist_ts_compaction/index.js",
"types": "./dist_ts_compaction/index.d.ts"
},
"./mcp": {
"import": "./dist_ts_mcp/index.js",
"types": "./dist_ts_mcp/index.d.ts"
}
},
"author": "Task Venture Capital GmbH",
@@ -37,11 +41,12 @@
"@types/node": "^25.6.0"
},
"dependencies": {
"@push.rocks/smartai": "^2.2.0",
"@modelcontextprotocol/sdk": "^1.29.0",
"@push.rocks/smartai": "^4.0.0",
"@push.rocks/smartfs": "^1.5.1",
"@push.rocks/smartrequest": "^5.0.1",
"@push.rocks/smartshell": "^3.3.8",
"ai": "^6.0.170",
"ai": "^6.0.182",
"zod": "^4.4.1"
},
"packageManager": "pnpm@10.28.2",
@@ -57,6 +62,7 @@
"ts/**/*",
"ts_tools/**/*",
"ts_compaction/**/*",
"ts_mcp/**/*",
"dist/**/*",
"dist_*/**/*",
"assets/**/*",
+629 -68
View File
File diff suppressed because it is too large Load Diff
+1 -1
View File
@@ -24,7 +24,7 @@ Each exports a factory returning a flat `ToolSet` (Record<string, Tool>):
4. **jsonTool()**`json_validate`, `json_transform`
## Key Dependencies
- `@push.rocks/smartai` ^2.0.0 — provider registry, `getModel()`, re-exports `tool`, `jsonSchema`
- `@push.rocks/smartai` ^4.0.0 — provider registry, `getModel()`, re-exports `tool`, `jsonSchema`
- `ai` ^6.0.0 — Vercel AI SDK v6 (`streamText`, `stepCountIs`, `ModelMessage`, `ToolSet`)
- `zod` ^3.25.0 — tool input schema definitions
- `@push.rocks/smartfs`, `smartshell`, `smartrequest` — tool implementations
+100 -3
View File
@@ -49,7 +49,7 @@ const result = await runAgent({
console.log(result.text); // "7 + 35 = 42"
console.log(result.steps); // number of agentic steps taken
console.log(result.usage); // { promptTokens, completionTokens, totalTokens }
console.log(result.usage); // { inputTokens, outputTokens, totalTokens, cacheReadTokens, cacheWriteTokens }
```
## Architecture
@@ -76,7 +76,8 @@ console.log(result.usage); // { promptTokens, completionTokens, totalTokens }
-**Parallel tool execution** — multiple tool calls in a single step are executed concurrently
- 🔧 **Auto-retry with backoff** — handles 429/529/503 errors with header-aware retry delays
- 🩹 **Tool call repair** — case-insensitive name matching + invalid tool sink prevents crashes
- 📊 **Token streaming**`onToken` and `onToolCall` callbacks for real-time progress
- 💰 **Provider prompt caching** — uses SmartAI cache helpers for Anthropic breakpoints and OpenAI cache affinity
- 📊 **Token and reasoning streaming**`onToken`, `onReasoning*`, and `onToolCall` callbacks for real-time progress
- 💥 **Context overflow handling** — detects overflow and invokes your `onContextOverflow` callback
## Core API
@@ -92,10 +93,15 @@ The single entry point. Options:
| `system` | `string` | `undefined` | System prompt |
| `tools` | `ToolSet` | `{}` | Tools the agent can call |
| `providerOptions` | `ProviderOptions` | `undefined` | Provider-specific AI SDK request options passed through to `streamText()` |
| `sessionId` | `string` | `undefined` | Stable session id used as provider prompt-cache affinity key where supported |
| `cache` | `'auto' \| false \| IAgentCacheOptions` | `'auto'` | Prompt-cache policy. Set `false` to disable SmartAgent cache defaults |
| `maxSteps` | `number` | `20` | Max agentic steps before stopping |
| `messages` | `ModelMessage[]` | `[]` | Conversation history (for multi-turn) |
| `maxRetries` | `number` | `5` | Max retries on rate-limit/server errors |
| `onToken` | `(delta: string) => void` | — | Streaming token callback |
| `onReasoningStart` | `(id: string) => void` | — | Called when a reasoning summary starts |
| `onReasoningDelta` | `(id: string, delta: string) => void` | — | Called for streamed reasoning summary text |
| `onReasoningEnd` | `(id: string, text: string) => void` | — | Called when a reasoning summary completes |
| `onToolCall` | `(name: string) => void` | — | Called when a tool is invoked |
| `onToolResult` | `(name: string, result: unknown) => void` | — | Called when a tool finishes |
| `validateCompletion` | `(result) => string \| void` | — | Return a string to reject and reprompt an incomplete run |
@@ -114,6 +120,8 @@ interface IAgentRunResult {
inputTokens: number;
outputTokens: number;
totalTokens: number;
cacheReadTokens: number;
cacheWriteTokens: number;
};
toolCalls: Array<{
toolName: string;
@@ -126,7 +134,7 @@ interface IAgentRunResult {
### OpenAI Provider Options
Use `providerOptions` for provider-specific request settings such as GPT reasoning effort. SmartAgent forwards the object unchanged to AI SDK `streamText()`.
Use `providerOptions` for provider-specific request settings such as GPT reasoning effort. SmartAgent merges cache defaults first, then applies your `providerOptions` so explicit caller options win.
```typescript
import { getModelSetup } from '@push.rocks/smartai';
@@ -157,6 +165,31 @@ const saved = result.toolCalls.some((call) =>
);
```
### Prompt Caching
SmartAgent enables prompt-cache defaults by default:
- Cache behavior is provided by `@push.rocks/smartai`, so provider-specific cache metadata stays centralized there.
- Anthropic-compatible models get cache breakpoints on the first two system messages and the two most recent non-system messages.
- OpenAI models get `store: false` by default and, when `sessionId` is provided, `promptCacheKey: sessionId` with `promptCacheRetention: 'in_memory'`.
- Longer retention is opt-in. Use `cache: { retention: '24h' }` for OpenAI or `cache: { retention: '1h' }` for Anthropic.
- Set `cache: false` to disable these defaults for a run.
```typescript
const result = await runAgent({
model,
sessionId: 'stable-session-id',
prompt: 'Continue the task.',
tools,
});
const noCache = await runAgent({
model,
prompt: 'One-off request.',
cache: false,
});
```
### Completion Validation
Use `validateCompletion` when a workflow must not finish unless a required side-effect happened. Return `void` to accept the run, or return a string to append that string as a new user message and continue. If retries are exhausted, `runAgent()` throws.
@@ -211,6 +244,70 @@ await runAgent({
});
```
## MCP Tools
MCP support lives in the `@push.rocks/smartagent/mcp` subpath. The main `@push.rocks/smartagent` import stays unchanged unless you opt in.
```typescript
import { runAgent } from '@push.rocks/smartagent';
import { createMcpTools } from '@push.rocks/smartagent/mcp';
const mcp = await createMcpTools({
servers: {
filesystem: {
type: 'stdio',
command: 'my-mcp-filesystem-server',
args: ['/workspace/project'],
},
},
});
try {
const result = await runAgent({
model,
prompt: 'Use the available MCP tools to inspect the project.',
tools: mcp.tools,
maxSteps: 10,
});
console.log(result.text);
} finally {
await mcp.close();
}
```
`createMcpTools()` supports `stdio`, Streamable HTTP, custom transports, and pre-created MCP clients. When multiple servers are configured, exposed tool names are prefixed with the sanitized server name, for example `filesystem__read_file`. The returned `toolNameMap` maps exposed AI SDK tool names back to their original MCP server/tool names.
## Reusable Tool Contexts
SmartAgent can build tools once and execute them through a host-provided context. The same shell, filesystem, and browser tool schemas can target local Node.js, SSH, MCP, or another transport supplied by the host app.
```typescript
import { createFilesystemTools, createShellTools, type IToolExecutionContext } from '@push.rocks/smartagent';
const context: IToolExecutionContext = {
cwd: '/workspace/project',
requestPermission: async (request) => {
// Host app decides whether to allow writes, commands, browser actions, etc.
},
shell: {
run: async (command, options) => sshRun(command, options),
},
fs: {
readFile: async (path, options) => sshRead(path, options),
writeFile: async (path, content) => sshWrite(path, content),
listDirectory: async (path, options) => sshList(path, options),
},
};
const tools = {
...createShellTools(context),
...createFilesystemTools(context, { includeDelete: false }),
};
```
For local execution, use `createLocalToolExecutionContext()` or the compatibility wrappers `shellTool()` and `filesystemTool()`.
## ToolRegistry
A lightweight helper for collecting tools:
+143
View File
@@ -0,0 +1,143 @@
import { expect, tap } from '@git.zone/tstest/tapbundle';
import {
createMcpTools,
formatMcpToolResult,
sanitizeMcpToolName,
type IMcpClientLike,
} from '../ts_mcp/index.js';
interface IFakeMcpClientState {
calls: Array<{ name: string; arguments?: Record<string, unknown> }>;
closed: boolean;
}
const createFakeMcpClient = (
toolName: string,
responsePrefix: string,
state: IFakeMcpClientState,
): IMcpClientLike => ({
listTools: async () => ({
tools: [
{
name: toolName,
description: `Runs ${toolName}`,
inputSchema: {
type: 'object',
properties: {
value: { type: 'string' },
},
required: ['value'],
},
},
],
} as any),
callTool: async (params) => {
state.calls.push(params);
return {
content: [
{
type: 'text',
text: `${responsePrefix}: ${String(params.arguments?.value ?? '')}`,
},
],
structuredContent: {
ok: true,
},
} as any;
},
close: async () => {
state.closed = true;
},
});
tap.test('sanitizeMcpToolName should produce AI SDK-safe tool names', async () => {
expect(sanitizeMcpToolName('read file/path')).toEqual('read_file_path');
expect(sanitizeMcpToolName('123-start')).toEqual('_123-start');
expect(sanitizeMcpToolName('***', 'fallback')).toEqual('fallback');
});
tap.test('createMcpTools should expose prefixed tools for multiple MCP servers', async () => {
const firstState: IFakeMcpClientState = { calls: [], closed: false };
const secondState: IFakeMcpClientState = { calls: [], closed: false };
const mcpTools = await createMcpTools({
servers: {
'server one': {
type: 'client',
client: createFakeMcpClient('echo tool', 'first', firstState),
},
'server/two': {
type: 'client',
client: createFakeMcpClient('echo tool', 'second', secondState),
},
},
});
expect(Object.keys(mcpTools.tools).sort()).toEqual([
'server_one__echo_tool',
'server_two__echo_tool',
]);
expect(mcpTools.toolNameMap.server_one__echo_tool).toEqual({
serverName: 'server one',
toolName: 'echo tool',
});
const output = await (mcpTools.tools.server_one__echo_tool as any).execute({ value: 'hello' });
expect(output).toInclude('first: hello');
expect(output).toInclude('Structured content');
expect(firstState.calls).toEqual([
{
name: 'echo tool',
arguments: { value: 'hello' },
},
]);
await mcpTools.close();
expect(firstState.closed).toBeTrue();
expect(secondState.closed).toBeTrue();
});
tap.test('createMcpTools should support paginated tool discovery and custom formatting', async () => {
const calls: Array<{ name: string; arguments?: Record<string, unknown> }> = [];
const client: IMcpClientLike = {
listTools: async (params) => ({
tools: [
{
name: params?.cursor ? 'second tool' : 'first tool',
inputSchema: { type: 'object' },
},
],
nextCursor: params?.cursor ? undefined : 'next',
} as any),
callTool: async (params) => {
calls.push(params);
return { content: [{ type: 'text', text: 'ok' }] } as any;
},
};
const mcpTools = await createMcpTools({
servers: {
only: { type: 'client', client },
},
formatResult: (_result, context) => context.exposedToolName,
});
expect(Object.keys(mcpTools.tools).sort()).toEqual(['first_tool', 'second_tool']);
const output = await (mcpTools.tools.second_tool as any).execute('ignored');
expect(output).toEqual('second_tool');
expect(calls).toEqual([
{
name: 'second tool',
arguments: {},
},
]);
});
tap.test('formatMcpToolResult should handle MCP result shapes', async () => {
expect(formatMcpToolResult({ toolResult: { ok: true } } as any)).toEqual({ ok: true });
expect(formatMcpToolResult({ content: [{ type: 'text', text: 'failed' }], isError: true } as any)).toInclude(
'MCP tool returned an error',
);
});
export default tap.start();
+191 -1
View File
@@ -1,7 +1,7 @@
import { expect, tap } from '@git.zone/tstest/tapbundle';
import { MockLanguageModelV3, convertArrayToReadableStream } from 'ai/test';
import * as smartagent from '../ts/index.js';
import { filesystemTool, shellTool, httpTool, jsonTool, truncateOutput } from '../ts_tools/index.js';
import { createBrowserTools, createFilesystemTools, createShellTools, filesystemTool, shellTool, httpTool, jsonTool, truncateOutput } from '../ts_tools/index.js';
import { compactMessages } from '../ts_compaction/index.js';
const createUsage = (inputTokens: number, outputTokens: number) => ({
@@ -33,6 +33,25 @@ const createTextStreamResult = (text: string) => ({
] as any[]),
});
const createReasoningStreamResult = (reasoning: string, text: string) => ({
stream: convertArrayToReadableStream([
{ type: 'stream-start', warnings: [] },
{ type: 'response-metadata', id: 'response-1', timestamp: new Date(0), modelId: 'mock-model' },
{ type: 'reasoning-start', id: 'reasoning-1' },
{ type: 'reasoning-delta', id: 'reasoning-1', delta: reasoning.slice(0, 7) },
{ type: 'reasoning-delta', id: 'reasoning-1', delta: reasoning.slice(7) },
{ type: 'reasoning-end', id: 'reasoning-1' },
{ type: 'text-start', id: 'text-1' },
{ type: 'text-delta', id: 'text-1', delta: text },
{ type: 'text-end', id: 'text-1' },
{
type: 'finish',
finishReason: { unified: 'stop', raw: 'stop' },
usage: createUsage(2, 2),
},
] as any[]),
});
const createToolCallStreamResult = (toolName: string, input: unknown) => ({
stream: convertArrayToReadableStream([
{ type: 'stream-start', warnings: [] },
@@ -104,6 +123,97 @@ tap.test('runAgent should forward providerOptions to streamText', async () => {
expect((model.doStreamCalls[0].providerOptions as any).openai.reasoningEffort).toEqual('xhigh');
});
tap.test('runAgent should add OpenAI cache defaults when sessionId is provided', async () => {
const model = new MockLanguageModelV3({
provider: 'openai',
modelId: 'gpt-5',
doStream: async () => createTextStreamResult('ok') as any,
});
const result = await smartagent.runAgent({
model,
prompt: 'hello',
sessionId: 'session-123',
providerOptions: {
openai: {
reasoningEffort: 'high',
},
} as any,
});
const openaiOptions = (model.doStreamCalls[0].providerOptions as any).openai;
expect(result.text).toEqual('ok');
expect(openaiOptions.store).toEqual(false);
expect(openaiOptions.promptCacheKey).toEqual('session-123');
expect(openaiOptions.promptCacheRetention).toEqual('in_memory');
expect(openaiOptions.reasoningEffort).toEqual('high');
});
tap.test('runAgent should stream reasoning summary callbacks', async () => {
const reasoningEvents: string[] = [];
const tokenDeltas: string[] = [];
const model = new MockLanguageModelV3({
doStream: async () => createReasoningStreamResult('thinking through it', 'done') as any,
});
const result = await smartagent.runAgent({
model,
prompt: 'hello',
onToken: (delta) => tokenDeltas.push(delta),
onReasoningStart: (id) => reasoningEvents.push('start:' + id),
onReasoningDelta: (id, delta) => reasoningEvents.push('delta:' + id + ':' + delta),
onReasoningEnd: (id, text) => reasoningEvents.push('end:' + id + ':' + text),
});
expect(result.text).toEqual('done');
expect(tokenDeltas.join('')).toEqual('done');
expect(reasoningEvents).toEqual([
'start:reasoning-1',
'delta:reasoning-1:thinkin',
'delta:reasoning-1:g through it',
'end:reasoning-1:thinking through it',
]);
});
tap.test('runAgent should mark Anthropic prompt cache breakpoints by default', async () => {
const model = new MockLanguageModelV3({
provider: 'anthropic',
modelId: 'claude-sonnet-4-5-20250929',
doStream: async () => createTextStreamResult('ok') as any,
});
const result = await smartagent.runAgent({
model,
system: 'stable system prompt',
prompt: 'hello',
});
const prompt = model.doStreamCalls[0].prompt as any[];
const systemMessage = prompt.find((message) => message.role === 'system');
const userMessage = prompt.find((message) => message.role === 'user');
expect(result.text).toEqual('ok');
expect(systemMessage.providerOptions?.anthropic?.cacheControl?.type).toEqual('ephemeral');
expect(userMessage.providerOptions?.anthropic?.cacheControl?.type).toEqual('ephemeral');
});
tap.test('runAgent should allow cache defaults to be disabled', async () => {
const model = new MockLanguageModelV3({
provider: 'openai',
modelId: 'gpt-5',
doStream: async () => createTextStreamResult('ok') as any,
});
await smartagent.runAgent({
model,
prompt: 'hello',
sessionId: 'session-123',
cache: false,
});
expect(model.doStreamCalls[0].providerOptions).toBeUndefined();
});
tap.test('runAgent should return final tool call records', async () => {
let streamCallCount = 0;
const callbackToolCalls: Array<{ name: string; input: unknown }> = [];
@@ -247,6 +357,29 @@ tap.test('shellTool returns expected tool names', async () => {
expect(names).toContain('run_command');
});
tap.test('createShellTools should execute through supplied context', async () => {
const permissions: unknown[] = [];
const calls: unknown[] = [];
const tools = createShellTools({
cwd: '/workspace',
requestPermission: async (request) => {
permissions.push(request);
},
shell: {
run: async (command, options) => {
calls.push({ command, options });
return { exitCode: 0, stdout: 'context-output', stderr: '' };
},
},
});
const result = await (tools.run_command as any).execute({ command: 'echo test', timeoutMs: 1234 });
expect(result).toEqual('context-output');
expect(calls[0]).toEqual({ command: 'echo test', options: { cwd: '/workspace', timeoutMs: 1234, abortSignal: undefined } });
expect(JSON.stringify(permissions[0])).toInclude('echo test');
});
tap.test('httpTool returns expected tool names', async () => {
const tools = httpTool();
const names = Object.keys(tools);
@@ -261,6 +394,63 @@ tap.test('jsonTool returns expected tool names', async () => {
expect(names).toContain('json_transform');
});
tap.test('createFilesystemTools should execute through supplied context', async () => {
const permissions: unknown[] = [];
const files = new Map<string, string>([['hello.txt', 'line1\nline2\nline3']]);
const tools = createFilesystemTools({
requestPermission: async (request) => {
permissions.push(request);
},
fs: {
readFile: async (filePath, options) => {
const content = files.get(filePath) ?? '';
if (options?.startLine || options?.endLine) {
const lines = content.split('\n');
return lines.slice((options.startLine ?? 1) - 1, options.endLine ?? lines.length).join('\n');
}
return content;
},
writeFile: async (filePath, content) => {
files.set(filePath, content);
return 'written';
},
listDirectory: async () => [...files.keys()],
},
}, { includeDelete: false });
const readResult = await (tools.read_file as any).execute({ path: 'hello.txt', startLine: 2, endLine: 2 });
const writeResult = await (tools.write_file as any).execute({ path: 'created.txt', content: 'created' });
const listResult = await (tools.list_directory as any).execute({ path: '.' });
expect(readResult).toEqual('line2');
expect(writeResult).toEqual('written');
expect(listResult).toInclude('created.txt');
expect(Object.keys(tools)).not.toContain('delete_file');
expect(JSON.stringify(permissions[0])).toInclude('created.txt');
});
tap.test('createBrowserTools should execute through supplied browser context', async () => {
const permissions: unknown[] = [];
const calls: unknown[] = [];
const tools = createBrowserTools({
requestPermission: async (request) => {
permissions.push(request);
},
browser: {
execute: async (input, options) => {
calls.push({ input, options });
return `browser:${input.action}:${input.url ?? ''}`;
},
},
});
const result = await (tools.browser as any).execute({ action: 'navigate', url: 'https://example.com', timeoutMs: 500 });
expect(result).toEqual('browser:navigate:https://example.com');
expect(JSON.stringify(calls[0])).toInclude('navigate');
expect(JSON.stringify(permissions[0])).toInclude('https://example.com');
});
tap.test('json_validate tool should validate valid JSON', async () => {
const tools = jsonTool();
const result = await (tools.json_validate as any).execute({
+1 -1
View File
@@ -3,6 +3,6 @@
*/
export const commitinfo = {
name: '@push.rocks/smartagent',
version: '3.1.0',
version: '3.7.0',
description: 'Agentic loop for ai-sdk (Vercel AI SDK). Wraps streamText with stopWhen for parallel multi-step tool execution. Built on @push.rocks/smartai.'
}
+38 -1
View File
@@ -3,7 +3,44 @@ export { ToolRegistry } from './smartagent.classes.toolregistry.js';
export { truncateOutput } from './smartagent.utils.truncation.js';
export type { ITruncateResult } from './smartagent.utils.truncation.js';
export { ContextOverflowError } from './smartagent.interfaces.js';
export type { IAgentRunOptions, IAgentRunResult, IAgentToolCallRecord, ProviderOptions } from './smartagent.interfaces.js';
export type {
IAgentCacheOptions,
IAgentRunOptions,
IAgentRunResult,
IAgentToolCallRecord,
ProviderOptions,
TAgentCacheRetention,
TAgentCacheSetting,
} from './smartagent.interfaces.js';
export {
createBrowserTools,
createFilesystemTools,
createLocalToolExecutionContext,
createShellTools,
filesystemTool,
formatShellResult,
formatToolOutput,
shellTool,
} from '../ts_tools/index.js';
export type {
IBrowserToolInput,
ICreateBrowserToolsOptions,
ICreateFilesystemToolsOptions,
ICreateShellToolsOptions,
IFilesystemToolOptions,
ILocalToolExecutionContextOptions,
IShellToolOptions,
IToolBrowserContext,
IToolExecutionContext,
IToolFilesystemContext,
IToolFilesystemListOptions,
IToolFilesystemReadOptions,
IToolPermissionRequest,
IToolRunOptions,
IToolShellContext,
IToolShellResult,
TBrowserToolAction,
} from '../ts_tools/index.js';
// Re-export tool() and z so consumers can define tools without extra imports
export { tool, jsonSchema } from '@push.rocks/smartai';
+23 -5
View File
@@ -4,9 +4,9 @@ import * as path from 'path';
export { path };
// ai-sdk core
import { streamText, generateText, stepCountIs } from 'ai';
import { streamText, generateText, stepCountIs, wrapLanguageModel } from 'ai';
export { streamText, generateText, stepCountIs };
export { streamText, generateText, stepCountIs, wrapLanguageModel };
export type {
ModelMessage,
@@ -15,11 +15,29 @@ export type {
} from 'ai';
// @push.rocks/smartai
import { tool, jsonSchema } from '@push.rocks/smartai';
import {
applySmartAiCacheProviderOptions,
createSmartAiCachingMiddleware,
jsonSchema,
resolveSmartAiCacheProvider,
tool,
} from '@push.rocks/smartai';
export { tool, jsonSchema };
export {
applySmartAiCacheProviderOptions,
createSmartAiCachingMiddleware,
resolveSmartAiCacheProvider,
tool,
jsonSchema,
};
export type { LanguageModelV3, TSmartAiProviderOptions as ProviderOptions } from '@push.rocks/smartai';
export type {
ISmartAiCacheOptions,
LanguageModelV3,
TSmartAiCacheRetention,
TSmartAiCacheSetting,
TSmartAiProviderOptions as ProviderOptions,
} from '@push.rocks/smartai';
// zod
import { z } from 'zod';
+85 -6
View File
@@ -90,6 +90,28 @@ function errorToString(error: unknown): string {
return String(error);
}
function tokenTotal(tokenUsage: unknown): number {
if (typeof tokenUsage === 'number') return tokenUsage;
if (tokenUsage && typeof tokenUsage === 'object' && typeof (tokenUsage as any).total === 'number') {
return (tokenUsage as any).total;
}
return 0;
}
function tokenCacheRead(tokenUsage: unknown): number {
if (tokenUsage && typeof tokenUsage === 'object' && typeof (tokenUsage as any).cacheRead === 'number') {
return (tokenUsage as any).cacheRead;
}
return 0;
}
function tokenCacheWrite(tokenUsage: unknown): number {
if (tokenUsage && typeof tokenUsage === 'object' && typeof (tokenUsage as any).cacheWrite === 'number') {
return (tokenUsage as any).cacheWrite;
}
return 0;
}
function recordToolCall(
toolCalls: IAgentToolCallRecord[],
toolCallIndexes: Map<string, number>,
@@ -129,11 +151,35 @@ export async function runAgent(options: IAgentRunOptions): Promise<IAgentRunResu
let attempt = 0;
let totalInput = 0;
let totalOutput = 0;
let totalCacheRead = 0;
let totalCacheWrite = 0;
let validationRetries = 0;
const toolCalls: IAgentToolCallRecord[] = [];
const toolCallIndexes = new Map<string, number>();
const reasoningTextById = new Map<string, string>();
const tools = options.tools ?? {};
const cache = options.cache ?? 'auto';
const configuredCacheProvider = typeof cache === 'object' ? cache.provider : undefined;
const messageCacheProvider = cache === false
? undefined
: configuredCacheProvider ?? plugins.resolveSmartAiCacheProvider(options.model.provider, options.model.modelId);
const model = messageCacheProvider
? plugins.wrapLanguageModel({
model: options.model,
middleware: plugins.createSmartAiCachingMiddleware({
...(typeof cache === 'object' ? cache : {}),
provider: messageCacheProvider,
}),
}) as unknown as plugins.LanguageModelV3
: options.model;
const providerOptions = plugins.applySmartAiCacheProviderOptions({
provider: options.model.provider,
modelId: options.model.modelId,
providerOptions: options.providerOptions,
cache,
sessionId: options.sessionId,
});
// Add a no-op sink for repaired-but-unrecognised tool calls
const allTools: plugins.ToolSet = {
@@ -157,11 +203,11 @@ export async function runAgent(options: IAgentRunOptions): Promise<IAgentRunResu
while (true) {
try {
const result = plugins.streamText({
model: options.model,
model,
system: options.system,
messages,
tools: allTools,
providerOptions: options.providerOptions,
providerOptions,
stopWhen: plugins.stepCountIs(options.maxSteps ?? 20),
maxRetries: 0, // handled manually below
abortSignal: options.abort,
@@ -182,8 +228,33 @@ export async function runAgent(options: IAgentRunOptions): Promise<IAgentRunResu
},
onChunk: ({ chunk }) => {
if (chunk.type === 'text-delta' && options.onToken) {
options.onToken((chunk as any).textDelta ?? (chunk as any).text ?? '');
const chunkType = String((chunk as any).type || '');
if (chunkType === 'text-delta' && options.onToken) {
options.onToken((chunk as any).delta ?? (chunk as any).textDelta ?? (chunk as any).text ?? '');
return;
}
if (chunkType === 'reasoning-start') {
const id = (chunk as any).id || 'reasoning';
reasoningTextById.set(id, '');
options.onReasoningStart?.(id, (chunk as any).providerMetadata);
return;
}
if (chunkType === 'reasoning-delta') {
const id = (chunk as any).id || 'reasoning';
const delta = (chunk as any).delta ?? (chunk as any).textDelta ?? (chunk as any).text ?? '';
if (!reasoningTextById.has(id)) {
reasoningTextById.set(id, '');
options.onReasoningStart?.(id, (chunk as any).providerMetadata);
}
reasoningTextById.set(id, (reasoningTextById.get(id) ?? '') + delta);
options.onReasoningDelta?.(id, delta, (chunk as any).providerMetadata);
return;
}
if (chunkType === 'reasoning-end') {
const id = (chunk as any).id || 'reasoning';
const text = reasoningTextById.get(id) ?? '';
reasoningTextById.delete(id);
options.onReasoningEnd?.(id, text, (chunk as any).providerMetadata);
}
},
@@ -218,8 +289,10 @@ export async function runAgent(options: IAgentRunOptions): Promise<IAgentRunResu
onStepFinish: ({ usage, toolCalls: stepToolCalls, toolResults, content }) => {
stepCount++;
totalInput += usage?.inputTokens ?? 0;
totalOutput += usage?.outputTokens ?? 0;
totalInput += tokenTotal((usage as any)?.inputTokens);
totalOutput += tokenTotal((usage as any)?.outputTokens);
totalCacheRead += tokenCacheRead((usage as any)?.inputTokens);
totalCacheWrite += tokenCacheWrite((usage as any)?.inputTokens);
for (const toolCall of stepToolCalls) {
recordToolCall(toolCalls, toolCallIndexes, toolCall);
}
@@ -239,6 +312,10 @@ export async function runAgent(options: IAgentRunOptions): Promise<IAgentRunResu
const finishReason = await result.finishReason;
const responseData = await result.response;
const responseMessages = responseData.messages as plugins.ModelMessage[];
for (const [id, reasoningText] of reasoningTextById) {
options.onReasoningEnd?.(id, reasoningText);
reasoningTextById.delete(id);
}
attempt = 0; // reset on success
@@ -251,6 +328,8 @@ export async function runAgent(options: IAgentRunOptions): Promise<IAgentRunResu
inputTokens: totalInput,
outputTokens: totalOutput,
totalTokens: totalInput + totalOutput,
cacheReadTokens: totalCacheRead,
cacheWriteTokens: totalCacheWrite,
},
toolCalls,
};
+29 -2
View File
@@ -1,6 +1,17 @@
import type { ToolSet, ModelMessage, LanguageModelV3, ProviderOptions } from './plugins.js';
import type {
ISmartAiCacheOptions,
ToolSet,
ModelMessage,
LanguageModelV3,
ProviderOptions,
TSmartAiCacheRetention,
TSmartAiCacheSetting,
} from './plugins.js';
export type { ProviderOptions };
export type IAgentCacheOptions = ISmartAiCacheOptions;
export type TAgentCacheRetention = TSmartAiCacheRetention;
export type TAgentCacheSetting = TSmartAiCacheSetting;
export interface IAgentToolCallRecord {
toolName: string;
@@ -20,6 +31,10 @@ export interface IAgentRunOptions {
tools?: ToolSet;
/** Provider-specific AI SDK request options passed through to streamText() */
providerOptions?: ProviderOptions;
/** Stable session id used as provider prompt-cache affinity key where supported. */
sessionId?: string;
/** Prompt-cache policy. Default: 'auto'. Set false to disable smartagent cache defaults. */
cache?: TAgentCacheSetting;
/**
* Maximum number of LLM↔tool round trips.
* Each step may execute multiple tools in parallel.
@@ -30,6 +45,12 @@ export interface IAgentRunOptions {
messages?: ModelMessage[];
/** Called for each streamed text delta */
onToken?: (delta: string) => void;
/** Called when the model starts a streamed reasoning summary */
onReasoningStart?: (id: string, providerMetadata?: unknown) => void;
/** Called for each streamed reasoning summary delta */
onReasoningDelta?: (id: string, delta: string, providerMetadata?: unknown) => void;
/** Called when a streamed reasoning summary completes */
onReasoningEnd?: (id: string, text: string, providerMetadata?: unknown) => void;
/** Called when a tool call starts */
onToolCall?: (toolName: string, input: unknown) => void;
/** Called when a tool call completes */
@@ -61,7 +82,13 @@ export interface IAgentRunResult {
/** Finish reason from the final step */
finishReason: string;
/** Accumulated token usage across all steps */
usage: { inputTokens: number; outputTokens: number; totalTokens: number };
usage: {
inputTokens: number;
outputTokens: number;
totalTokens: number;
cacheReadTokens: number;
cacheWriteTokens: number;
};
/** Tool calls observed during the run, including inputs and outputs/errors when available */
toolCalls: IAgentToolCallRecord[];
}
+344
View File
@@ -0,0 +1,344 @@
import { Client, type ClientOptions } from '@modelcontextprotocol/sdk/client/index.js';
import {
StdioClientTransport,
type StdioServerParameters,
} from '@modelcontextprotocol/sdk/client/stdio.js';
import {
StreamableHTTPClientTransport,
type StreamableHTTPClientTransportOptions,
} from '@modelcontextprotocol/sdk/client/streamableHttp.js';
import type { RequestOptions } from '@modelcontextprotocol/sdk/shared/protocol.js';
import type { Transport } from '@modelcontextprotocol/sdk/shared/transport.js';
import { jsonSchema, tool } from '@push.rocks/smartai';
import type { ToolSet } from '@push.rocks/smartai';
export type TMcpListToolsResult = Awaited<ReturnType<Client['listTools']>>;
export type TMcpToolMetadata = TMcpListToolsResult['tools'][number];
export type TMcpCallToolResult = Awaited<ReturnType<Client['callTool']>>;
export interface IMcpClientLike {
listTools(params?: { cursor?: string }, options?: RequestOptions): Promise<TMcpListToolsResult>;
callTool(
params: { name: string; arguments?: Record<string, unknown> },
resultSchema?: unknown,
options?: RequestOptions,
): Promise<TMcpCallToolResult>;
close?(): Promise<void>;
}
export interface IMcpServerClientOptions {
clientName?: string;
clientVersion?: string;
clientOptions?: ClientOptions;
requestOptions?: RequestOptions;
}
export interface IMcpStdioServerConfig extends StdioServerParameters, IMcpServerClientOptions {
type: 'stdio';
}
export interface IMcpStreamableHttpServerConfig extends IMcpServerClientOptions {
type: 'streamableHttp' | 'http';
url: string | URL;
transportOptions?: StreamableHTTPClientTransportOptions;
}
export interface IMcpTransportServerConfig extends IMcpServerClientOptions {
type: 'transport';
transport: Transport;
}
export interface IMcpClientServerConfig {
type: 'client';
client: IMcpClientLike;
}
export type TMcpServerConfig =
| IMcpStdioServerConfig
| IMcpStreamableHttpServerConfig
| IMcpTransportServerConfig
| IMcpClientServerConfig;
export interface IMcpToolFormatContext {
serverName: string;
toolName: string;
exposedToolName: string;
}
export interface ICreateMcpToolsOptions {
servers: Record<string, TMcpServerConfig>;
clientName?: string;
clientVersion?: string;
clientOptions?: ClientOptions;
requestOptions?: RequestOptions;
listToolsRequestOptions?: RequestOptions;
callToolRequestOptions?: RequestOptions;
prefixToolNames?: boolean;
formatResult?: (result: TMcpCallToolResult, context: IMcpToolFormatContext) => unknown;
}
export interface IMcpConnectedServer {
name: string;
client: IMcpClientLike;
tools: TMcpToolMetadata[];
}
export interface IMcpToolNameMapping {
serverName: string;
toolName: string;
}
export interface ICreateMcpToolsResult {
tools: ToolSet;
servers: IMcpConnectedServer[];
toolNameMap: Record<string, IMcpToolNameMapping>;
close: () => Promise<void>;
}
export const createMcpTools = async (
options: ICreateMcpToolsOptions,
): Promise<ICreateMcpToolsResult> => {
const serverEntries = Object.entries(options.servers);
if (serverEntries.length === 0) {
throw new Error('createMcpTools requires at least one MCP server.');
}
const connectedServers: IMcpConnectedServer[] = [];
try {
for (const [serverName, config] of serverEntries) {
const client = await connectMcpServer(serverName, config, options);
const tools = await listAllMcpTools(
client,
options.listToolsRequestOptions ?? getServerRequestOptions(config) ?? options.requestOptions,
);
connectedServers.push({ name: serverName, client, tools });
}
const exposedTools: ToolSet = {};
const toolNameMap: Record<string, IMcpToolNameMapping> = {};
const usedToolNames = new Set<string>();
const formatResult = options.formatResult ?? formatMcpToolResult;
const shouldPrefixToolNames = options.prefixToolNames ?? connectedServers.length > 1;
for (const server of connectedServers) {
for (const mcpTool of server.tools) {
const baseToolName = createExposedToolName(server.name, mcpTool.name, shouldPrefixToolNames);
const exposedToolName = createUniqueToolName(baseToolName, usedToolNames);
usedToolNames.add(exposedToolName);
toolNameMap[exposedToolName] = {
serverName: server.name,
toolName: mcpTool.name,
};
exposedTools[exposedToolName] = tool({
description: mcpTool.description ?? `MCP tool ${mcpTool.name} from ${server.name}`,
inputSchema: jsonSchema(mcpTool.inputSchema as never),
execute: async (input: unknown) => {
const result = await server.client.callTool(
{
name: mcpTool.name,
arguments: normalizeMcpArguments(input),
},
undefined,
options.callToolRequestOptions ?? options.requestOptions,
);
return formatResult(result, {
serverName: server.name,
toolName: mcpTool.name,
exposedToolName,
});
},
});
}
}
return {
tools: exposedTools,
servers: connectedServers,
toolNameMap,
close: async () => closeMcpClients(connectedServers),
};
} catch (error) {
await closeMcpClients(connectedServers);
throw error;
}
};
export const sanitizeMcpToolName = (name: string, fallback = 'mcp_tool'): string => {
const sanitized = name
.trim()
.replace(/[^a-zA-Z0-9_-]+/g, '_')
.replace(/^_+|_+$/g, '');
const safeName = sanitized || fallback;
return /^[a-zA-Z_]/.test(safeName) ? safeName : `_${safeName}`;
};
export const formatMcpToolResult = (result: TMcpCallToolResult): unknown => {
const resultRecord = result as Record<string, unknown>;
if (Object.prototype.hasOwnProperty.call(resultRecord, 'toolResult')) {
return resultRecord.toolResult;
}
const content = Array.isArray(resultRecord.content)
? resultRecord.content.map(formatMcpContentPart).filter(Boolean).join('\n\n')
: '';
const formattedContent = resultRecord.isError && content
? `MCP tool returned an error:\n${content}`
: content;
const structuredContent = resultRecord.structuredContent;
if (structuredContent !== undefined && formattedContent) {
return `${formattedContent}\n\nStructured content:\n${stringifyJson(structuredContent)}`;
}
if (structuredContent !== undefined) {
return structuredContent;
}
if (formattedContent) {
return formattedContent;
}
return stringifyJson(result);
};
const connectMcpServer = async (
serverName: string,
config: TMcpServerConfig,
options: ICreateMcpToolsOptions,
): Promise<IMcpClientLike> => {
if (config.type === 'client') {
return config.client;
}
const client = new Client(
{
name: config.clientName ?? options.clientName ?? 'smartagent-mcp',
version: config.clientVersion ?? options.clientVersion ?? '1.0.0',
},
config.clientOptions ?? options.clientOptions ?? { capabilities: {} },
);
const transport = createMcpTransport(config);
await client.connect(transport, config.requestOptions ?? options.requestOptions);
return client;
};
const createMcpTransport = (
config: Exclude<TMcpServerConfig, IMcpClientServerConfig>,
): Transport => {
if (config.type === 'stdio') {
const { type, clientName, clientVersion, clientOptions, requestOptions, ...serverParameters } = config;
return new StdioClientTransport(serverParameters);
}
if (config.type === 'streamableHttp' || config.type === 'http') {
return new StreamableHTTPClientTransport(new URL(config.url), config.transportOptions);
}
if (config.type === 'transport') {
return config.transport;
}
throw new Error('Unsupported MCP server transport.');
};
const getServerRequestOptions = (config: TMcpServerConfig): RequestOptions | undefined => {
return config.type === 'client' ? undefined : config.requestOptions;
};
const listAllMcpTools = async (
client: IMcpClientLike,
requestOptions?: RequestOptions,
): Promise<TMcpToolMetadata[]> => {
const tools: TMcpToolMetadata[] = [];
let cursor: string | undefined;
do {
const result = await client.listTools(cursor ? { cursor } : undefined, requestOptions);
tools.push(...result.tools);
cursor = result.nextCursor;
} while (cursor);
return tools;
};
const createExposedToolName = (
serverName: string,
toolName: string,
prefixToolName: boolean,
): string => {
const safeToolName = sanitizeMcpToolName(toolName);
if (!prefixToolName) {
return safeToolName;
}
return `${sanitizeMcpToolName(serverName, 'mcp')}__${safeToolName}`;
};
const createUniqueToolName = (baseToolName: string, usedToolNames: Set<string>): string => {
if (!usedToolNames.has(baseToolName)) {
return baseToolName;
}
let suffix = 2;
while (usedToolNames.has(`${baseToolName}_${suffix}`)) {
suffix++;
}
return `${baseToolName}_${suffix}`;
};
const normalizeMcpArguments = (input: unknown): Record<string, unknown> => {
if (!input || typeof input !== 'object' || Array.isArray(input)) {
return {};
}
return input as Record<string, unknown>;
};
const formatMcpContentPart = (part: unknown): string => {
if (!part || typeof part !== 'object') {
return stringifyJson(part);
}
const partRecord = part as Record<string, unknown>;
const type = partRecord.type;
if (type === 'text') {
return typeof partRecord.text === 'string' ? partRecord.text : '';
}
if (type === 'image' || type === 'audio') {
const data = typeof partRecord.data === 'string' ? partRecord.data : '';
const mimeType = typeof partRecord.mimeType === 'string' ? partRecord.mimeType : 'application/octet-stream';
return `[${type}: ${mimeType}; base64 length ${data.length}]`;
}
if (type === 'resource') {
return formatMcpResourcePart(partRecord.resource);
}
if (type === 'resource_link') {
const name = typeof partRecord.name === 'string' ? partRecord.name : 'resource';
const uri = typeof partRecord.uri === 'string' ? partRecord.uri : '';
return `Resource link ${name}${uri ? `: ${uri}` : ''}`;
}
return stringifyJson(part);
};
const formatMcpResourcePart = (resource: unknown): string => {
if (!resource || typeof resource !== 'object') {
return stringifyJson(resource);
}
const resourceRecord = resource as Record<string, unknown>;
const uri = typeof resourceRecord.uri === 'string' ? resourceRecord.uri : 'resource';
if (typeof resourceRecord.text === 'string') {
return `Resource ${uri}:\n${resourceRecord.text}`;
}
if (typeof resourceRecord.blob === 'string') {
const mimeType = typeof resourceRecord.mimeType === 'string' ? resourceRecord.mimeType : 'application/octet-stream';
return `[resource ${uri}: ${mimeType}; base64 length ${resourceRecord.blob.length}]`;
}
return stringifyJson(resource);
};
const stringifyJson = (value: unknown): string => {
if (typeof value === 'string') {
return value;
}
try {
return JSON.stringify(value, undefined, 2);
} catch {
return String(value);
}
};
const closeMcpClients = async (servers: IMcpConnectedServer[]): Promise<void> => {
await Promise.all(servers.map((server) => server.client.close?.()));
};
+21 -4
View File
@@ -1,7 +1,24 @@
export { filesystemTool } from './tool.filesystem.js';
export type { IFilesystemToolOptions } from './tool.filesystem.js';
export { shellTool } from './tool.shell.js';
export type { IShellToolOptions } from './tool.shell.js';
export { createBrowserTools } from './tool.browser.js';
export type { ICreateBrowserToolsOptions } from './tool.browser.js';
export { createFilesystemTools, filesystemTool } from './tool.filesystem.js';
export type { ICreateFilesystemToolsOptions, IFilesystemToolOptions } from './tool.filesystem.js';
export { createShellTools, shellTool } from './tool.shell.js';
export type { ICreateShellToolsOptions, IShellToolOptions } from './tool.shell.js';
export { createLocalToolExecutionContext, formatShellResult, formatToolOutput } from './tool.context.js';
export type {
IBrowserToolInput,
ILocalToolExecutionContextOptions,
IToolBrowserContext,
IToolExecutionContext,
IToolFilesystemContext,
IToolFilesystemListOptions,
IToolFilesystemReadOptions,
IToolPermissionRequest,
IToolRunOptions,
IToolShellContext,
IToolShellResult,
TBrowserToolAction,
} from './tool.context.js';
export { httpTool } from './tool.http.js';
export { jsonTool } from './tool.json.js';
export { truncateOutput } from './plugins.js';
+4 -3
View File
@@ -1,8 +1,9 @@
// node native
import * as path from 'path';
import * as fs from 'fs';
import * as childProcess from 'node:child_process';
import * as fs from 'node:fs';
import * as path from 'node:path';
export { path, fs };
export { childProcess, fs, path };
// zod
import { z } from 'zod';
+94
View File
@@ -0,0 +1,94 @@
import * as plugins from './plugins.js';
import {
formatToolOutput,
type IBrowserToolInput,
type IToolExecutionContext,
type TBrowserToolAction,
} from './tool.context.js';
export interface ICreateBrowserToolsOptions {
/** Maximum output lines before truncating. */
maxLines?: number;
/** Maximum output bytes before truncating. */
maxBytes?: number;
}
const browserActions = new Set<TBrowserToolAction>(['navigate', 'snapshot', 'screenshot', 'click', 'fill', 'press', 'evaluate', 'close']);
export function createBrowserTools(context: IToolExecutionContext, options: ICreateBrowserToolsOptions = {}): plugins.ToolSet {
return {
browser: plugins.tool({
description: [
'Control a browser supplied by the host execution context for web UI inspection and interaction.',
'Actions: navigate, snapshot, screenshot, click, fill, press, evaluate, close.',
'Use snapshot after navigation or interaction to inspect page text and interactive selectors before choosing the next action.',
'Actions that navigate or modify page state require host permission when configured.',
].join(' '),
inputSchema: plugins.z.object({
action: plugins.z.string().default('snapshot').describe('Action: navigate, snapshot, screenshot, click, fill, press, evaluate, or close'),
url: plugins.z.string().optional().describe('URL for navigate'),
selector: plugins.z.string().optional().describe('CSS or Playwright selector for click/fill'),
text: plugins.z.string().optional().describe('Text for fill, key name for press, or screenshot mode/full-page hint'),
script: plugins.z.string().optional().describe('JavaScript expression or function body for evaluate'),
timeoutMs: plugins.z.number().optional().describe('Optional action timeout in milliseconds'),
}),
execute: async (input: IBrowserToolInput) => {
if (!context.browser) {
throw new Error('Browser tool is not available in this execution context.');
}
const action = normalizeBrowserAction(input.action);
await requestBrowserPermission(context, { ...input, action });
const result = await context.browser.execute({ ...input, action }, {
timeoutMs: input.timeoutMs,
abortSignal: context.abortSignal,
});
return plugins.truncateOutput(formatToolOutput(result), {
maxLines: options.maxLines,
maxBytes: options.maxBytes,
}).content;
},
}),
};
}
const normalizeBrowserAction = (input: unknown): TBrowserToolAction => {
const action = typeof input === 'string' && input.trim()
? input.trim().toLowerCase()
: 'snapshot';
if (browserActions.has(action as TBrowserToolAction)) {
return action as TBrowserToolAction;
}
throw new Error(`Unsupported browser action: ${String(input)}. Use one of: ${[...browserActions].join(', ')}.`);
};
const requestBrowserPermission = async (context: IToolExecutionContext, input: IBrowserToolInput & { action: TBrowserToolAction }): Promise<void> => {
if (!context.requestPermission) return;
if (input.action === 'snapshot' || input.action === 'screenshot') return;
const titleByAction: Record<TBrowserToolAction, string> = {
navigate: 'Navigate browser',
snapshot: 'Inspect browser',
screenshot: 'Capture browser screenshot',
click: 'Click browser element',
fill: 'Fill browser element',
press: 'Press browser key',
evaluate: 'Evaluate browser JavaScript',
close: 'Close browser session',
};
await context.requestPermission({
type: 'browser',
title: titleByAction[input.action],
metadata: {
action: input.action,
url: input.url,
selector: input.selector,
key: input.action === 'press' ? input.text : undefined,
textLength: input.action === 'fill' ? input.text?.length ?? 0 : undefined,
scriptPreview: input.action === 'evaluate' && input.script ? compactMetadataText(input.script) : undefined,
},
});
};
const compactMetadataText = (text: string): string => {
const compacted = text.replace(/\s+/g, ' ').trim();
return compacted.length > 160 ? `${compacted.slice(0, 157)}...` : compacted;
};
+207
View File
@@ -0,0 +1,207 @@
import * as plugins from './plugins.js';
export interface IToolPermissionRequest {
type: string;
title: string;
metadata?: Record<string, unknown>;
}
export interface IToolRunOptions {
cwd?: string;
timeoutMs?: number;
abortSignal?: AbortSignal;
}
export interface IToolShellResult {
exitCode: number;
stdout: string;
stderr?: string;
signal?: string;
}
export interface IToolShellContext {
run(command: string, options?: IToolRunOptions): Promise<IToolShellResult | string>;
}
export interface IToolFilesystemReadOptions {
startLine?: number;
endLine?: number;
}
export interface IToolFilesystemListOptions {
recursive?: boolean;
}
export interface IToolFilesystemContext {
readFile(filePath: string, options?: IToolFilesystemReadOptions): Promise<string>;
writeFile(filePath: string, content: string): Promise<string | void>;
listDirectory(directoryPath: string, options?: IToolFilesystemListOptions): Promise<string[] | string>;
deletePath?(targetPath: string): Promise<string | void>;
}
export type TBrowserToolAction = 'navigate' | 'snapshot' | 'screenshot' | 'click' | 'fill' | 'press' | 'evaluate' | 'close';
export interface IBrowserToolInput {
action?: TBrowserToolAction | string;
url?: string;
selector?: string;
text?: string;
script?: string;
timeoutMs?: number;
}
export interface IToolBrowserContext {
execute(input: IBrowserToolInput, options?: { timeoutMs?: number; abortSignal?: AbortSignal }): Promise<unknown>;
}
export interface IToolExecutionContext {
cwd?: string;
rootDir?: string;
abortSignal?: AbortSignal;
shell?: IToolShellContext;
fs?: IToolFilesystemContext;
browser?: IToolBrowserContext;
requestPermission?: (request: IToolPermissionRequest) => Promise<void>;
}
export interface ILocalToolExecutionContextOptions {
cwd?: string;
rootDir?: string;
abortSignal?: AbortSignal;
requestPermission?: (request: IToolPermissionRequest) => Promise<void>;
}
export const createLocalToolExecutionContext = (options: ILocalToolExecutionContextOptions = {}): IToolExecutionContext => {
const cwd = options.cwd ?? process.cwd();
const rootDir = options.rootDir;
return {
cwd,
rootDir,
abortSignal: options.abortSignal,
requestPermission: options.requestPermission,
shell: {
run: (command, runOptions) => runLocalShellCommand(command, {
cwd: resolveLocalPath(runOptions?.cwd ?? cwd, rootDir),
timeoutMs: runOptions?.timeoutMs,
abortSignal: runOptions?.abortSignal ?? options.abortSignal,
}),
},
fs: {
readFile: async (filePath, readOptions) => {
const resolved = resolveLocalPath(filePath, rootDir, cwd);
const content = await plugins.fs.promises.readFile(resolved, 'utf8');
if (readOptions?.startLine !== undefined || readOptions?.endLine !== undefined) {
const lines = content.split('\n');
const start = Math.max((readOptions.startLine ?? 1) - 1, 0);
const end = Math.max(readOptions.endLine ?? lines.length, start);
return lines.slice(start, end).join('\n');
}
return content;
},
writeFile: async (filePath, content) => {
const resolved = resolveLocalPath(filePath, rootDir, cwd);
await plugins.fs.promises.mkdir(plugins.path.dirname(resolved), { recursive: true });
await plugins.fs.promises.writeFile(resolved, content, 'utf8');
return `Written ${Buffer.byteLength(content, 'utf8')} bytes to ${filePath}`;
},
listDirectory: async (directoryPath, listOptions) => {
const resolved = resolveLocalPath(directoryPath, rootDir, cwd);
return listLocalDirectory(resolved, !!listOptions?.recursive);
},
deletePath: async (targetPath) => {
const resolved = resolveLocalPath(targetPath, rootDir, cwd);
await plugins.fs.promises.rm(resolved, { recursive: false, force: false });
return `Deleted ${targetPath}`;
},
},
};
};
export const formatShellResult = (result: IToolShellResult | string): string => {
if (typeof result === 'string') return result;
if (result.exitCode === 0) return result.stdout;
return [
`Exit code: ${result.exitCode}`,
result.signal ? `Signal: ${result.signal}` : '',
`stdout:\n${result.stdout}`,
`stderr:\n${result.stderr ?? ''}`,
].filter(Boolean).join('\n');
};
export const formatToolOutput = (output: unknown): string => {
if (typeof output === 'string') return output;
try {
return JSON.stringify(output, undefined, 2);
} catch {
return String(output);
}
};
const resolveLocalPath = (targetPath: string, rootDir?: string, baseDir?: string): string => {
const base = rootDir ?? baseDir ?? process.cwd();
const resolved = plugins.path.isAbsolute(targetPath)
? plugins.path.resolve(targetPath)
: plugins.path.resolve(base, targetPath);
if (rootDir) {
const resolvedRoot = plugins.path.resolve(rootDir);
if (resolved !== resolvedRoot && !resolved.startsWith(resolvedRoot + plugins.path.sep)) {
throw new Error(`Access denied: "${targetPath}" is outside allowed root "${rootDir}"`);
}
}
return resolved;
};
const listLocalDirectory = async (directoryPath: string, recursive: boolean): Promise<string[]> => {
const entries = await plugins.fs.promises.readdir(directoryPath, { withFileTypes: true });
const result: string[] = [];
for (const entry of entries.sort((a, b) => a.name.localeCompare(b.name))) {
const relativePath = entry.name + (entry.isDirectory() ? '/' : '');
result.push(relativePath);
if (recursive && entry.isDirectory()) {
const childEntries = await listLocalDirectory(plugins.path.join(directoryPath, entry.name), true);
result.push(...childEntries.map((childEntry) => `${entry.name}/${childEntry}`));
}
}
return result;
};
const runLocalShellCommand = async (command: string, options: IToolRunOptions): Promise<IToolShellResult> => {
return new Promise<IToolShellResult>((resolve) => {
const child = plugins.childProcess.spawn('bash', ['-lc', command], {
cwd: options.cwd,
stdio: ['ignore', 'pipe', 'pipe'],
});
let stdout = '';
let stderr = '';
let timedOut = false;
const timeout = options.timeoutMs && options.timeoutMs > 0
? setTimeout(() => {
timedOut = true;
child.kill('SIGTERM');
}, options.timeoutMs)
: undefined;
const abort = () => child.kill('SIGTERM');
options.abortSignal?.addEventListener('abort', abort, { once: true });
child.stdout?.on('data', (chunk) => {
stdout += chunk.toString();
});
child.stderr?.on('data', (chunk) => {
stderr += chunk.toString();
});
child.on('error', (error) => {
if (timeout) clearTimeout(timeout);
options.abortSignal?.removeEventListener('abort', abort);
resolve({ exitCode: 1, stdout, stderr: `${stderr}${error.message}` });
});
child.on('close', (code, signal) => {
if (timeout) clearTimeout(timeout);
options.abortSignal?.removeEventListener('abort', abort);
resolve({
exitCode: code ?? (timedOut ? 124 : 1),
stdout,
stderr: timedOut ? `${stderr}\nCommand timed out after ${options.timeoutMs}ms.`.trim() : stderr,
signal: signal ?? undefined,
});
});
});
};
+80 -78
View File
@@ -1,30 +1,43 @@
import * as plugins from './plugins.js';
import {
createLocalToolExecutionContext,
formatToolOutput,
type IToolExecutionContext,
} from './tool.context.js';
export interface IFilesystemToolOptions {
/** Restrict file access to this directory. Default: process.cwd() */
rootDir?: string;
/** Execution context. Defaults to a local Node.js context. */
context?: IToolExecutionContext;
/** Include delete_file. Default: true for compatibility. */
includeDelete?: boolean;
/** Maximum output lines before truncating. */
maxLines?: number;
/** Maximum output bytes before truncating. */
maxBytes?: number;
}
function validatePath(filePath: string, rootDir?: string): string {
const resolved = plugins.path.resolve(filePath);
if (rootDir) {
const resolvedRoot = plugins.path.resolve(rootDir);
if (!resolved.startsWith(resolvedRoot + plugins.path.sep) && resolved !== resolvedRoot) {
throw new Error(`Access denied: "${filePath}" is outside allowed root "${rootDir}"`);
}
}
return resolved;
export interface ICreateFilesystemToolsOptions {
/** Include delete_file. Default: true. */
includeDelete?: boolean;
/** Maximum output lines before truncating. */
maxLines?: number;
/** Maximum output bytes before truncating. */
maxBytes?: number;
}
export function filesystemTool(options?: IFilesystemToolOptions): plugins.ToolSet {
const rootDir = options?.rootDir;
return {
export function createFilesystemTools(context: IToolExecutionContext, options: ICreateFilesystemToolsOptions = {}): plugins.ToolSet {
const truncate = (output: unknown) => plugins.truncateOutput(formatToolOutput(output), {
maxLines: options.maxLines,
maxBytes: options.maxBytes,
}).content;
const tools: plugins.ToolSet = {
read_file: plugins.tool({
description:
'Read file contents. Returns the full text or a specified line range.',
'Read a UTF-8 file in the active workspace. Paths may be absolute or relative to the workspace root.',
inputSchema: plugins.z.object({
path: plugins.z.string().describe('Absolute path to the file'),
path: plugins.z.string().describe('File path'),
startLine: plugins.z
.number()
.optional()
@@ -43,89 +56,78 @@ export function filesystemTool(options?: IFilesystemToolOptions): plugins.ToolSe
startLine?: number;
endLine?: number;
}) => {
const resolved = validatePath(filePath, rootDir);
const content = plugins.fs.readFileSync(resolved, 'utf-8');
if (startLine !== undefined || endLine !== undefined) {
const lines = content.split('\n');
const start = (startLine ?? 1) - 1;
const end = endLine ?? lines.length;
const sliced = lines.slice(start, end).join('\n');
return plugins.truncateOutput(sliced).content;
if (!context.fs) {
throw new Error('Filesystem tools are not available in this execution context.');
}
return plugins.truncateOutput(content).content;
return truncate(await context.fs.readFile(filePath, { startLine, endLine }));
},
}),
write_file: plugins.tool({
description:
'Write content to a file (creates parent dirs if needed, overwrites existing).',
'Write UTF-8 content to a file in the active workspace. Creates parent directories and overwrites existing content. Requires host permission when configured.',
inputSchema: plugins.z.object({
path: plugins.z.string().describe('Absolute path to the file'),
content: plugins.z.string().describe('Content to write'),
path: plugins.z.string().describe('File path'),
content: plugins.z.string().describe('Complete file content to write'),
}),
execute: async ({ path: filePath, content }: { path: string; content: string }) => {
const resolved = validatePath(filePath, rootDir);
const dir = plugins.path.dirname(resolved);
plugins.fs.mkdirSync(dir, { recursive: true });
plugins.fs.writeFileSync(resolved, content, 'utf-8');
return `Written ${content.length} characters to ${filePath}`;
if (!context.fs) {
throw new Error('Filesystem tools are not available in this execution context.');
}
await context.requestPermission?.({
type: 'write',
title: 'Write file',
metadata: { path: filePath, bytes: Buffer.byteLength(content, 'utf8') },
});
const result = await context.fs.writeFile(filePath, content);
return truncate(result ?? `Written ${Buffer.byteLength(content, 'utf8')} bytes to ${filePath}`);
},
}),
list_directory: plugins.tool({
description: 'List files and directories at the given path.',
description: 'List files and directories in the active workspace. Paths may be absolute or relative to the workspace root.',
inputSchema: plugins.z.object({
path: plugins.z.string().describe('Directory path to list'),
path: plugins.z.string().default('.').describe('Directory path to list'),
recursive: plugins.z
.boolean()
.optional()
.describe('List recursively (default: false)'),
.describe('List recursively. Default: false'),
}),
execute: async ({
path: dirPath,
recursive,
}: {
path: string;
recursive?: boolean;
}) => {
const resolved = validatePath(dirPath, rootDir);
function listDir(dir: string, prefix: string = ''): string[] {
const entries = plugins.fs.readdirSync(dir, { withFileTypes: true });
const result: string[] = [];
for (const entry of entries) {
const rel = prefix ? `${prefix}/${entry.name}` : entry.name;
const indicator = entry.isDirectory() ? '/' : '';
result.push(`${rel}${indicator}`);
if (recursive && entry.isDirectory()) {
result.push(...listDir(plugins.path.join(dir, entry.name), rel));
}
}
return result;
execute: async ({ path: directoryPath, recursive }: { path: string; recursive?: boolean }) => {
if (!context.fs) {
throw new Error('Filesystem tools are not available in this execution context.');
}
const entries = listDir(resolved);
return plugins.truncateOutput(entries.join('\n')).content;
},
}),
delete_file: plugins.tool({
description: 'Delete a file or empty directory.',
inputSchema: plugins.z.object({
path: plugins.z.string().describe('Path to delete'),
}),
execute: async ({ path: filePath }: { path: string }) => {
const resolved = validatePath(filePath, rootDir);
const stat = plugins.fs.statSync(resolved);
if (stat.isDirectory()) {
plugins.fs.rmdirSync(resolved);
} else {
plugins.fs.unlinkSync(resolved);
}
return `Deleted ${filePath}`;
const result = await context.fs.listDirectory(directoryPath, { recursive });
return truncate(Array.isArray(result) ? result.join('\n') : result);
},
}),
};
if (options.includeDelete !== false) {
tools.delete_file = plugins.tool({
description: 'Delete a file or empty directory in the active workspace. Requires host permission when configured.',
inputSchema: plugins.z.object({
path: plugins.z.string().describe('Path to delete'),
}),
execute: async ({ path: targetPath }: { path: string }) => {
if (!context.fs?.deletePath) {
throw new Error('Deleting files is not available in this execution context.');
}
await context.requestPermission?.({
type: 'delete',
title: 'Delete file',
metadata: { path: targetPath },
});
const result = await context.fs.deletePath(targetPath);
return truncate(result ?? `Deleted ${targetPath}`);
},
});
}
return tools;
}
export function filesystemTool(options?: IFilesystemToolOptions): plugins.ToolSet {
const context = options?.context ?? createLocalToolExecutionContext({ rootDir: options?.rootDir });
return createFilesystemTools(context, options);
}
+51 -17
View File
@@ -1,19 +1,37 @@
import * as plugins from './plugins.js';
import {
createLocalToolExecutionContext,
formatShellResult,
type IToolExecutionContext,
} from './tool.context.js';
export interface IShellToolOptions {
/** Allowed commands whitelist. If empty, all commands are allowed. */
allowedCommands?: string[];
/** Working directory for shell execution */
cwd?: string;
/** Execution context. Defaults to a local Node.js context. */
context?: IToolExecutionContext;
/** Maximum output lines before truncating. */
maxLines?: number;
/** Maximum output bytes before truncating. */
maxBytes?: number;
}
export function shellTool(options?: IShellToolOptions): plugins.ToolSet {
const smartshell = new plugins.smartshell.Smartshell({ executor: 'bash' });
export interface ICreateShellToolsOptions {
/** Allowed commands whitelist. If empty, all commands are allowed. */
allowedCommands?: string[];
/** Maximum output lines before truncating. */
maxLines?: number;
/** Maximum output bytes before truncating. */
maxBytes?: number;
}
export function createShellTools(context: IToolExecutionContext, options: ICreateShellToolsOptions = {}): plugins.ToolSet {
return {
run_command: plugins.tool({
description:
'Execute a shell command. Provide the full command string. stdout and stderr are returned.',
'Execute a shell command in the active workspace. Provide the full command string. stdout and stderr are returned.',
inputSchema: plugins.z.object({
command: plugins.z.string().describe('The shell command to execute'),
cwd: plugins.z
@@ -24,39 +42,55 @@ export function shellTool(options?: IShellToolOptions): plugins.ToolSet {
.number()
.optional()
.describe('Timeout in milliseconds'),
timeoutMs: plugins.z
.number()
.optional()
.describe('Timeout in milliseconds'),
}),
execute: async ({
command,
cwd,
timeout,
timeoutMs,
}: {
command: string;
cwd?: string;
timeout?: number;
timeoutMs?: number;
}) => {
// Validate against allowed commands whitelist
if (options?.allowedCommands?.length) {
if (!context.shell) {
throw new Error('Shell tool is not available in this execution context.');
}
if (options.allowedCommands?.length) {
const baseCommand = command.split(/\s+/)[0];
if (!options.allowedCommands.includes(baseCommand)) {
return `Command "${baseCommand}" is not in the allowed commands list: ${options.allowedCommands.join(', ')}`;
}
}
// Build full command string with cd prefix if cwd specified
const effectiveCwd = cwd ?? options?.cwd;
const fullCommand = effectiveCwd
? `cd ${JSON.stringify(effectiveCwd)} && ${command}`
: command;
await context.requestPermission?.({
type: 'shell',
title: 'Run shell command',
metadata: { command, cwd: cwd ?? context.cwd },
});
const execResult = await smartshell.exec(fullCommand);
const execResult = await context.shell.run(command, {
cwd: cwd ?? context.cwd,
timeoutMs: timeoutMs ?? timeout,
abortSignal: context.abortSignal,
});
const output =
execResult.exitCode === 0
? execResult.stdout
: `Exit code: ${execResult.exitCode}\nstdout:\n${execResult.stdout}\nstderr:\n${execResult.stderr ?? ''}`;
return plugins.truncateOutput(output).content;
return plugins.truncateOutput(formatShellResult(execResult), {
maxLines: options.maxLines,
maxBytes: options.maxBytes,
}).content;
},
}),
};
}
export function shellTool(options?: IShellToolOptions): plugins.ToolSet {
const context = options?.context ?? createLocalToolExecutionContext({ cwd: options?.cwd });
return createShellTools(context, options);
}