Model Context Protocol (MCP) Server

repotographer includes an official Model Context Protocol (MCP) server over stdio. This allows coding assistants like Claude Desktop, Cursor, and OpenCode to query your GitHub ecosystem, propose domain groupings, and render diagrams on demand.

Starting the Server Manually

Run the server directly on the command line:

repotographer mcp

The process listens for JSON-RPC messages on standard input and outputs responses to standard output.

---

Generating Client Configurations

You can generate ready-to-use configuration blocks for your specific client with repotographer mcp config:

# Print configurations for all supported clients
repotographer mcp config

# Print configuration for OpenCode
repotographer mcp config --client opencode

# Print configuration for Claude Desktop with ambient environment variables
repotographer mcp config --client claude --env

---

Client Configurations

Add repotographer to your MCP client configuration file:

OpenCode (opencode.json)

{
  "mcp": {
    "repotographer": {
      "type": "local",
      "command": ["repotographer", "mcp"],
      "environment": {
        "GEMINI_API_KEY": "your-gemini-api-key"
      },
      "enabled": true
    }
  }
}

Claude Desktop (claude_desktop_config.json)

{
  "mcpServers": {
    "repotographer": {
      "command": "/usr/local/bin/repotographer",
      "args": ["mcp"],
      "env": {
        "GEMINI_API_KEY": "your-gemini-api-key"
      }
    }
  }
}

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "repotographer": {
      "command": "repotographer",
      "args": ["mcp"],
      "env": {
        "GEMINI_API_KEY": "your-gemini-api-key"
      }
    }
  }
}

Antigravity (.antigravity/mcp_config.json)

{
  "mcpServers": {
    "repotographer": {
      "command": "repotographer",
      "args": ["mcp"],
      "env": {
        "GEMINI_API_KEY": "your-gemini-api-key"
      }
    }
  }
}

---

Agent Plugin & Skills

repotographer includes an official bundle adhering to the Agent Plugins Specification:

• Plugin Manifest: plugins/repotographer/plugin.json
• MCP Declaration: plugins/repotographer/mcp.json
• Skill Workflow: plugins/repotographer/skills/repotographer/SKILL.md

Agent runtimes automatically discover these files to guide users through mapping, taxonomy curation, and rendering.

---

Available MCP Tools

The server exposes three tools implementing typed JSON schemas:

1. map_github_account

Fetches public repositories, detects technology connections, derives a domain taxonomy, and returns the complete graph structure.

Arguments:

• owner (string, required): GitHub username or organization name.
• account_type (string, optional): "user" or "org".
• limit (integer, optional): Maximum repositories to analyze (default 500).
• use_llm (boolean, optional): Refine taxonomy with Gemini (default true).
• model (string, optional): Gemini model name (default "gemini-3.7-flash").
• use_vertex (boolean, optional): Use Google Cloud Vertex AI backend.

2. suggest_taxonomy

Analyzes repositories and returns a proposed domain pillar taxonomy with project assignments without building full graph geometry.

Arguments:

• owner (string, required): GitHub username or organization.
• account_type (string, optional): "user" or "org".
• limit (integer, optional): Maximum repositories to analyze.
• model (string, optional): Gemini model name.
• use_vertex (boolean, optional): Use Google Cloud Vertex AI backend.

3. render_graph

Accepts a concept graph object and writes the requested formats (HTML, DOT, PNG, JSON) to disk.

Arguments:

• graph (object, required): The complete concept graph data structure.
• out_dir (string, optional): Directory path where files should be written.
• formats (array of strings, optional): ["html", "dot", "png", "json"].