mcp recipe

Wrap an MCP server as Agentmatic tools

Connect to any MCP (Model Context Protocol) server and expose its tools to your agent. Three lines for stdio servers, four for SSE.

3 min read · Published May 4, 2026 · Languages: python, typescript

The pattern

MCP is becoming the open standard for AI tools. Any MCP server (filesystem, git, browser, Postgres, GitHub) is one client connection away from being callable by your agent.

MCP in one line: An MCP server is a process that exposes tools over JSON-RPC; an MCP client (your agent) discovers and calls them.

Stdio transport

from agentmatic.tools.mcp import MCPClient

fs = MCPClient.stdio([
    "npx", "-y", "@modelcontextprotocol/server-filesystem", "/Users/me/projects"
])
tools = await fs.list_tools_as_agentmatic()

agent = create_react_agent(llm=OpenAI(), tools=tools)

SSE / HTTP transport

mcp = await MCPClient.sse("https://my-mcp.example.com/mcp")
tools = await mcp.list_tools_as_agentmatic()

TypeScript

import { MCPClient, createReactAgent, OpenAI } from '@agentmatic/core';

const fs = await MCPClient.stdio([
  'npx', '-y', '@modelcontextprotocol/server-filesystem', '/Users/me/projects'
]);
const tools = await fs.listToolsAsAgentmatic();
const agent = createReactAgent({ llm: new OpenAI(), tools });

Discover servers from mcp.json

If you already have an mcp.json for Claude Desktop / Cursor:

from agentmatic.tools.mcp import discover_from_config

clients = await discover_from_config()  # reads ~/.mcp/mcp.json + project-local
tools = await asyncio.gather(*[c.list_tools_as_agentmatic() for c in clients.values()])