agent-tools-mcp

Discover agent-callable resources from any MCP-compatible agent (Claude, Cursor, Cline, Continue, …).

Backed by agent-tools.cloud, an open directory that indexes three kinds of agent-callable resources:

A2A tells agents who can do the job. x402 lets them pay for it. MCP gives them the tools to do it.

This is a discovery layer, not a facilitator — your agent keeps full custody of payment. The tools never auto-pay; they only find resources and return call templates. Call the stats tool for live counts across all three protocols.

Also listed on Smithery for one-click hosted install.

Tools

Typical flow: start with search_all if you don't know which resource type fits, or go straight to the protocol-specific search (search / search_mcp_servers / search_agents), then call the matching get* tool for the full call template.

What agents get back (x402 example)

search() and get() return a normalised service record. For services hosted directly on agent-tools.cloud, every paid endpoint advertises the full x402 v2 extensions.bazaar block in its 402 challenge — including JSON Schema and a worked request body example — so the agent can call without trial-and-error:

// shape of get("agent-tools-cloud-signal-token")
{
  "slug": "agent-tools-cloud-signal-token",
  "endpoint": "https://agent-tools.cloud/v1/signal/token",
  "method": "POST",
  "price_usd": 0.01,
  "network": "base-mainnet",
  "asset": "USDC",
  "bazaar": {
    "info": {
      "input":  { "type": "http", "method": "POST", "bodyType": "json",
                  "body":   { "chain": "base", "token": "0x4200...0006" } },
      "output": { "type": "json",
                  "example": { "signal": "buy", "score": 0.78, "confidence": 0.62 } }
    },
    "schema": { "type": "object", "properties": { "body": { "properties":
                  { "chain": {"type":"string","enum":["base","ethereum","solana"]},
                    "token": {"type":"string"} },
                  "required": ["chain", "token"] } } }
  }
}

The response also passes through payment-required (challenge) and payment-response headers, both exposed via Access-Control-Expose-Headers so a browser-side fetch() (e.g. x402-fetch) can read them.

Third-party entries scraped from awesome-x402 / x402scan / x402.org, the MCP registry / PulseMCP, and public A2A agent-cards are passed through as-is and may or may not include full call metadata.

Quick Start

Claude Code CLI

claude mcp add agent-tools -- uvx agent-tools-mcp

Claude Desktop / Cursor / Cline

Add to your MCP config (~/.config/Claude/claude_desktop_config.json, ~/.cursor/mcp.json, …):

{
  "mcpServers": {
    "agent-tools": {
      "command": "uvx",
      "args": ["agent-tools-mcp"]
    }
  }
}

Remote (no install)

Most clients also accept a url-based remote MCP server (Streamable HTTP; the client must send Accept: application/json, text/event-stream):

{
  "mcpServers": {
    "agent-tools": { "url": "https://agent-tools.cloud/mcp-discovery/" }
  }
}

From source

pip install agent-tools-mcp        # or `uv tool install agent-tools-mcp`
agent-tools-mcp                    # stdio server, ready for an MCP client

Environment Variables

Debugging

# Probe with the official MCP Inspector
npx -y @modelcontextprotocol/inspector uvx agent-tools-mcp

# Or raw JSON-RPC
echo '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | uvx agent-tools-mcp

License

Apache-2.0. See LICENSE.

Related

• Directory site: https://agent-tools.cloud
• x402 spec: https://x402.org
• MCP spec: https://modelcontextprotocol.io
• A2A spec: https://a2a.dev