Introduction
Ganglia is an MCP server that gives AI coding agents a real understanding of your codebase — not file contents, but structure: who calls what, what breaks if you touch X, where a route ends up in the database.
It indexes your project into a call graph and exposes 45 semantic tools over the Model Context Protocol. Works with Claude Code, Cursor, Windsurf, and any MCP-compatible client.
Why it exists
Section titled “Why it exists”AI agents are fast. Too fast to know that processPayment is three hops from the function they just refactored. Ganglia gives them a nervous system.
| Problem | Ganglia’s answer |
|---|---|
| AI guesses which code is safe to touch | Real call graph — every caller, every callee |
| No blast-radius awareness before edits | code_impact with severity scores |
| Every session starts cold | Cross-session memory via code_memory |
| Critical paths invisible to AI | Danger zones + zone_guard enforcement |
| Full file reads waste tokens | Function-level reads — 80% fewer tokens |
How it works
Section titled “How it works”Ganglia parses your source into a graph (Cozo by default, FalkorDB optional). The MCP server exposes that graph as tools. Your AI agent calls the tools instead of reading files blindly.
AI Agent → MCP Client → Ganglia MCP Server → Code Graph → Your CodebaseNo cloud required. One binary. Runs alongside your editor.
Supported languages
Section titled “Supported languages”TypeScript · JavaScript · Rust · Python · Go (more coming)
Supported clients
Section titled “Supported clients”- Claude Code — add to
.mcp.jsonin your project root - Cursor — add to
.cursor/mcp.json - Windsurf — add to MCP config
- Any MCP-compatible client
Next steps
Section titled “Next steps”- Installation — get the binary and connect your client
- Quick Start — running in under 5 minutes
- Tools Overview — all 57 tools at a glance