TPMJS is an open-source registry and execution layer for AI agent tools. It continuously scans npm for packages tagged as AI tools, extracts their schemas, validates them, scores their quality, and makes them instantly callable by any AI agent through MCP, SDK, or API.
The problem: AI agents are only as useful as the tools they can access, but discovering, validating, and connecting tools is scattered across GitHub repos, blog posts, and Discord threads. There is no central place to find what exists, no standard way to know if a tool actually works, and wiring up each tool to each client is tedious manual config. TPMJS fixes this by treating AI tools the way npm treats packages — one registry, automatic quality checks, and a universal interface.
Publish an npm package with the tpmjs keyword and TPMJS picks it up within minutes. It extracts Zod/JSON schemas, runs health checks, computes quality scores, and serves the tool via MCP endpoints that work with Claude Code, Cursor, Windsurf, and any MCP-compatible client. You can group tools into collections, test them with auto-generated scenarios, and build agents that use them — all from one place.
Turborepo monorepo with Next.js (App Router), PostgreSQL via Prisma, deployed on Vercel. Tool execution runs in isolated Deno sandboxes on Railway. The CLI and SDK are published to npm under the @tpmjs scope. MCP implementation follows the Model Context Protocol specification with HTTP transport.
TPMJS was created by Ajax Davis (Thomas Davis) in 2024. It started as a side project to scratch an itch — finding usable AI tools was harder than it should be — and grew into a full platform. It is free to use and open source.