Configuration

Configuration

There are two sides to MCP configuration:

  1. Client config — The mcpServers JSON your AI app reads to connect to servers
  2. mcpm.json — The file you include in your project so the registry knows how to run it

Client Config Files

Client Config File
Claude Desktop claude_desktop_config.json
Cursor ~/.cursor/mcp.json
Windsurf ~/.codeium/windsurf/mcp_config.json
ChatGPT Settings → Integrations → MCP
VS Code / Copilot .vscode/mcp.json
Claude Code .mcp.json

Client Config Format

{
  "mcpServers": {
    "my-tool": {
      "command": "node",
      "args": ["server.js"],
      "env": {
        "API_KEY": "sk-..."
      }
    }
  }
}

mcpm.json — The Bridge

When you publish a package to mcpm.dev, include an mcpm.json file. Its mcp field maps 1:1 to the mcpServers entry above:

{
  "mcp": {
    "transport": "stdio",
    "command": "node",
    "args": ["server.js"],
    "env": {}
  }
}

When users run mcpm-dev add <your-tool>, the CLI reads your mcpm.json and generates the exact client config they need. Without it, the CLI falls back to a generic npx -y <name> command.
```