AI Tools Directory — MCP server

A remote MCP server exposing a curated index of 221 AI tools across 21 industry departments, so AI assistants can look up, filter and shortlist tools instead of guessing.

• Endpoint: https://ai-tools-mcp.toolboxes.top/mcp

• Transport: Streamable HTTP (JSON-RPC 2.0)

• Auth: none

• Source data: ai.toolboxes.top

Quick connect

Claude Desktop / Claude Code

{
  "mcpServers": {
    "ai-tools-directory": {
      "type": "http",
      "url": "https://ai-tools-mcp.toolboxes.top/mcp"
    }
  }
}

Cursor — .cursor/mcp.json

Same structure as above.

Any client

POST https://ai-tools-mcp.toolboxes.top/mcp
Content-Type: application/json

{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"my-client","version":"1.0"}}}

Related MCP server: HyperStore MCP

Tools

Resources

Example

{
  "jsonrpc": "2.0", "id": 2, "method": "tools/call",
  "params": { "name": "search_ai_tools", "arguments": { "department": "legal", "pricing": "freemium", "limit": 3 } }
}

**Harvey** — Legal · paid
...

Data quality — please read

• The directory is reliable. Name, URL, department, pricing tier, description and tags are human-maintained and reviewed.

• freeTier is NOT verified. It is regex-extracted from vendor pages and has known failure modes: some tools return a login wall instead of a pricing page, and the extractor can't always tell "free tier needs no card" from "paid plans need a card".

  • The server drops contradictory signals rather than guessing.

  • The server's instructions require any assistant using freeTier to tell the user it is auto-detected and may be outdated.

  • Coverage is ~87% of tools reachable; of those, about half yielded a usable signal.

• A missing freeTier field means "no data", not "no free tier."

If you find wrong free-tier data, please open an issue — that's exactly the feedback needed to improve the extractor.

Self-host

The server is a single Cloudflare Worker with no dependencies.

npm install
node build-dataset.mjs          # regenerate dataset.json (edit it, or point it at your own data)
npx wrangler deploy

Requires CLOUDFLARE_ACCOUNT_ID and CLOUDFLARE_API_TOKEN in the environment — do not commit them.

node test.mjs https://ai-tools-mcp.toolboxes.top   # 36 end-to-end assertions

To run locally:

npx wrangler dev --port 8791
node test.mjs http://127.0.0.1:8791

Notes on the deployment

• workers.dev hostnames are blocked by the GFW, so the canonical hostname is ai-tools-mcp.toolboxes.top (a Cloudflare custom domain on the Worker).

• The endpoint is served directly from a Worker; there is no origin server.

License

Code: MIT. Dataset: CC BY 4.0 — free to use with attribution to ai.toolboxes.top.