Skip to main content

Overview

Mirobody functions as an OAuth-enabled MCP server that exposes its built-in tools to AI clients like Claude Desktop, Cursor, and custom MCP clients. The endpoint speaks JSON-RPC 2.0 and is served by src/mcp/.
The Model Context Protocol (MCP) is a standard interface for AI applications to access tools, resources, and context.
The engine exposes two request paths:
The standalone binary listens on HTTP_HOST:HTTP_PORT, which defaults to 0.0.0.0:8080. Adjust the port below if you’ve changed HTTP_PORT.

MCP Configuration

For Claude Desktop & Cursor

Add Mirobody to your MCP configuration:
~/Library/Application Support/Claude/claude_desktop_config.json
Configuration file location:
For Cursor:

Restart Your Client

After updating the configuration:
1

Save configuration file

Ensure your changes are saved to the MCP config file
2

Restart Claude/Cursor

Close and reopen Claude Desktop or Cursor completely
3

Verify connection

The MCP server should appear in the client’s MCP servers listIn Claude Desktop: Click the MCP icon to see connected serversIn Cursor: Check MCP status in settings
4

Test a tool

Try using one of your tools:“How did I sleep recently?”The client should call the family_health tool. (A no-auth smoke test: ask it to echo some text.)
You should see your Mirobody tools available in the AI client!

Authentication Flow

When using tools through MCP clients, OAuth authentication is handled automatically:
MCP request flow: user -> client -> MCP -> Mirobody -> answer
The OAuth token is managed by Mirobody. Users authenticate once, and tools have secure access to their data.

Available Endpoints

The MCP server exposes the following endpoint: Base URL: http://localhost:8080/mcp (or http://localhost:8080/mcp/{secret} for a personal-MCP secret) Protocol: JSON-RPC 2.0 Methods include:
  • tools/list: List all available tools
  • tools/call: Execute a specific tool

For ChatGPT

Mirobody can also function as a GPT app, enabling capabilities within ChatGPT. Read more at ChatGPT Apps.

Custom MCP Clients

Direct HTTP Requests

You can call tools directly via HTTP:

Next Steps

Built-in Tools

Explore default health tools

Add Custom Tools

Create your own tools

MCP API Reference

Detailed MCP protocol documentation