MCP server
Graphify glossary
A server implementing the Model Context Protocol — the open standard that lets AI assistants call external tools, like a code graph.
An MCP server is a server implementing the Model Context Protocol, the open standard that lets AI assistants call external tools. Instead of every assistant inventing its own plugin system, MCP defines one interface: a server exposes named tools with typed inputs, and any MCP-capable assistant can discover and call them. It's the piece of plumbing that makes a tool assistant-agnostic.
Graphify ships an MCP server that exposes eight tools over the graph — querying structure, tracing paths, explaining nodes, and so on. It runs over stdio for the common single-assistant case, plus a shared HTTP mode when you want several assistants (or several sessions) hitting the same graph. Because the interface is standard, the same graph works natively across the 17 assistants Graphify supports — Claude Code, Cursor, Copilot, Codex, Gemini CLI, Aider, and others — without per-assistant integration work.
In practice you rarely configure it by hand: graphify install detects which assistants you use and wires the MCP server into them. From then on, when your assistant needs to know what calls a function or how two services connect, it calls a graph tool instead of grepping — and the graph stays on your machine, queried locally, with no telemetry.
The architectural point is worth noticing: MCP turns the graph into infrastructure rather than a feature of one editor. Build the graph once and it's the same memory in every assistant you work with — switch tools, keep the map. That's also why the graph survives assistant churn: the investment lives in the standard interface, not in any one vendor's plugin format.
Related terms
- Code graph — A knowledge graph built from a codebase: functions, classes, config, and docs become nodes; calls, imports, and references become edges.
- Token reduction — The context-window savings from answering with graph structure instead of pasting whole files into an assistant's context.