Your code graph, as native tools.

The 10 tools

Seven graph tools and three PR tools. Every tool also accepts an optional project_path for multi-project setups. Parameters for each are in the MCP tools reference.

• query_graph

  Search the graph with BFS or DFS; returns relevant nodes and edges as text context.

• get_node

  Full details for one node, by label or ID.

• get_neighbors

  All direct neighbors of a node, with edge details.

• shortest_path

  The shortest path between two concepts in the graph.

• get_community

  Every node in a community, by community ID.

• god_nodes

  The most connected nodes: the core abstractions of the codebase.

• graph_stats

  Summary statistics: node count, edge count, communities, confidence breakdown.

• list_prs

  Open GitHub PRs with CI status, review state, and graph impact per PR.

• get_pr_impact

  Detailed graph impact for one PR: files changed, communities affected, nodes touched.

• triage_prs

  All actionable open PRs with impact data, for review priority and merge order.

The server also exposes 6 read-only resources: the full graph report, graph stats, god nodes, surprising cross-community connections, a confidence audit, and suggested questions for the codebase.

Run the server

The server ships in the graphifyy package behind the mcp extra. Two transports: stdio (default) for a per-developer local server, and streamable HTTP for one shared process a whole team points at.

# install with the mcp extra, then serve the graph:$ uv tool install "graphifyy[mcp]"$ python -m graphify.serve graphify-out/graph.json # stdio is the default transport: one local server per developer

# one shared process for the whole team (streamable HTTP):$ python -m graphify.serve graphify-out/graph.json --transport http --port 8080 # expose beyond localhost: bind 0.0.0.0 AND set a key$ python -m graphify.serve graphify-out/graph.json --transport http --host 0.0.0.0 --api-key "$SECRET" # clients point at:  http://<host>:8080/mcp

HTTP flags: --host (default 127.0.0.1, loopback only), --port (8080), --path (/mcp), --api-key (or GRAPHIFY_API_KEY) for bearer-token auth, plus --stateless and --json-response for load-balanced or CI deployments.

Connect your assistant

All stdio configs are the same shape: run python -m graphify.serve with the path to your graph.json. Pick your client:

Claude Code

# one-liner:$ claude mcp add graphify -- python -m graphify.serve graphify-out/graph.json # or declaratively, in .mcp.json at the project root:{  "mcpServers": {    "graphify": {      "command": "python",      "args": ["-m", "graphify.serve", "graphify-out/graph.json"]    }  }}

Cursor

Add to .cursor/mcp.json in the project (or ~/.cursor/mcp.json globally):

{  "mcpServers": {    "graphify": {      "command": "python",      "args": ["-m", "graphify.serve", "graphify-out/graph.json"]    }  }}

Codex

Add to ~/.codex/config.toml:

[mcp_servers.graphify]command = "python"args = ["-m", "graphify.serve", "graphify-out/graph.json"]

Gemini CLI

Add to ~/.gemini/settings.json (or the project's .gemini/settings.json):

{  "mcpServers": {    "graphify": {      "command": "python",      "args": ["-m", "graphify.serve", "graphify-out/graph.json"]    }  }}

Any client, over HTTP

Running the shared HTTP server? Point any MCP client at the URL instead of a command:

{  "mcpServers": {    "graphify": {      "url": "http://localhost:8080/mcp"    }  }}

Related pages

The MCP tools referencedocuments every tool's parameters and the server flags in full. Separately, Graphify hosts a docs-search MCP server (one tool, search_graphify_docs) so your agent can search this documentation in-editor; that one is hosted and needs nothing installed. New to Graphify entirely? Start with what is Graphify or the tutorial.

Common questions