How to Use OpenClaw with MCP Servers — Tool Integration

MCP (Model Context Protocol) is how OpenClaw's tool ecosystem works under the hood. Understanding it lets you extend your agent with virtually any capability. Here's the practical guide.

What MCP Does

MCP servers expose tools and resources to your agent. Think of them as plugins — each server adds a set of things your agent can do. The agent automatically discovers what tools are available and uses them when relevant.

Connecting an MCP Server

openclaw mcp add github --path node
# Or for a remote server:
openclaw mcp add custom-api --url https://your-mcp-server.com

Common MCP Servers

File system access:

openclaw mcp add filesystem --path node --args "--root /path/to/allow"

Web search:

openclaw mcp add brave-search --env BRAVE_API_KEY=YOUR_KEY

Database access:

openclaw mcp add sqlite --args "/path/to/database.db"

Building a Custom MCP Server

If you have an internal API, write a simple MCP server:

// my-custom-mcp.js
import { Server } from '@modelcontextprotocol/sdk/server/index.js';

const server = new Server({ name: 'my-api', version: '1.0.0' });

server.setRequestHandler('tools/list', async () => ({
  tools: [{
    name: 'get_customer',
    description: 'Get customer data from internal CRM',
    inputSchema: {
      type: 'object',
      properties: { email: { type: 'string' } },
      required: ['email']
    }
  }]
}));

server.setRequestHandler('tools/call', async (request) => {
  if (request.params.name === 'get_customer') {
    const customer = await yourCRM.getCustomer(request.params.arguments.email);
    return { content: [{ type: 'text', text: JSON.stringify(customer) }] };
  }
});

openclaw mcp add my-api --path node --args "my-custom-mcp.js"

Listing Active MCP Servers

openclaw mcp list
# Shows: server name | status | tools available

Debugging MCP Issues

openclaw mcp logs github --last 20
# Common issue: server crashes on startup
# Check: correct path, dependencies installed, env vars set

The OpenClaw Playbook has an advanced MCP chapter covering tool design patterns, how to build MCP servers that work well with LLM agents, and a library of community MCP servers worth installing.