gh-claude-fatbrainmap-mcp

# gh-claude-fatbrainmap-mcp An MCP server that turns GitHub issue/PR relationship graphs into interactive mindmaps. Designed to pair with the [`gh-claude`](https://github.com/jsheffie/gh-claude) Claude Code plugin: when you run `/gh-claude:gh-related 42`, the skill builds a JSON graph and hands it to this server, which spins up a local web UI and returns a clickable URL. ## What you get - A FastMCP server that exposes one tool: `render_mindmap(graph)` - A bundled FastAPI web app that serves a React Flow canvas with two layout engines: - **Hierarchy** (elk, layered) — root issue at top, related items branching down - **Organic** (d3-force) — coggle-style floating nodes pulled together by edges - Bezier edges, solid for direct references and dashed for semantic matches - Drag, zoom, minimap, right-click to hide a node - Every node title links to the underlying GitHub issue/PR ## Install Until this is on PyPI, install from source: ```bash git clone https://github.com/jsheffie/gh-claude-fatbrainmap-mcp.git cd gh-claude-fatbrainmap-mcp python -m venv .venv && source .venv/bin/activate pip install -e . (cd ui && npm install && npm run build) # Register with Claude Code (user scope = available in every project) claude mcp add fatbrainmap --scope user -- "$(pwd)/.venv/bin/gh-claude-fatbrainmap-mcp" ``` Then restart Claude Code (or run `/mcp` to verify). Once connected, the `gh-claude` plugin's `/gh-claude:gh-related N` will detect the MCP tool and append a clickable mindmap URL to its output. ## Local development ```bash # Python deps python -m venv .venv && source .venv/bin/activate pip install -e '.[dev]' # UI deps cd ui && npm install && cd .. # Run both dev servers ./scripts/dev.sh # In another terminal: seed an example graph and open the URL it prints python scripts/seed.py ``` The MCP server proper (stdio transport for Claude Code) is launched via `gh-claude-fatbrainmap-mcp` (same name as the console script in `pyproject.toml`). ## Build for distribution ```bash ./scripts/build.sh # builds UI into server/static/ then python wheel into dist/ ``` ## JSON contract The skill must send graphs matching `server/schemas.py`: ```json { "version": 1, "repo": "owner/name", "root": 42, "nodes": [ { "id": 42, "kind": "issue", "state": "open", "title": "...", "url": "https://github.com/...", "labels": ["bug"], "milestone": "v2.1" } ], "edges": [ { "source": 42, "target": 38, "kind": "direct", "reason": "fixes #38" } ] } ``` See `examples/sample-graph.json` for a complete example. ## License MIT