toolpulse — agent tool reliability monitoring (TypeScript)

One-line wrapper. Catches the silent tool failures your agent can't.

npm install toolpulse

import { monitor } from "toolpulse";

const searchWeb = monitor(async (query: string) => {
  // your existing tool, unchanged
  return await fetch(`...?q=${query}`).then(r => r.json());
}, { toolName: "search_web", agentId: "my-agent" });

That's it. Every call is recorded, schema drift is detected, and you get a dashboard at toolpulse.io.

export TOOLPULSE_API_KEY=tp_live_xxxxxxxxxxxxx

Why ToolPulse

Agent tools fail silently — APIs change response shapes, latency creeps up, exceptions get caught and forgotten. ToolPulse records every tool call, fingerprints the response shape, alerts on schema drift, and runs synthetic health checks.

Drop-in for popular agent frameworks

// Vercel AI SDK
import { tool } from "ai";
import { monitor } from "toolpulse";

const search = tool({
  description: "search the web",
  parameters: z.object({ query: z.string() }),
  execute: monitor(async ({ query }) => { /* ... */ }, { toolName: "search" }),
});

// LangChain.js
import { DynamicStructuredTool } from "@langchain/core/tools";
new DynamicStructuredTool({
  name: "search",
  func: monitor(async (input) => { /* ... */ }, { toolName: "search" }),
  // ...
});

// MCP TypeScript server
server.tool("search", { query: z.string() }, monitor(handler, { toolName: "search" }));

License

MIT