AI agents & MCP

Role of MCP

Codegenome runs a local MCP server that exposes graph queries as tools. Agents call these tools instead of guessing structure from partial file reads.

Setup flow

1. Build the graph — codegenome analyze .
2. Start MCP — codegenome mcp-start or HTTP via legacy CLI
3. Install client config — python -m codegenome.installer …
4. Generate rules — codegenome rules --client cursor .
5. Restart your editor

Example tools

• search_nodes — find symbols by name or path
• get_neighbors — imports, callers, callees
• get_entry_points, get_dead_code, get_circular_deps
• get_god_nodes, get_complexity, get_churn

Full list: MCP tools reference.

Agent rules

Templates in the repository extensions/templates/ teach agents when to call MCP tools. Install with:

codegenome rules --client cursor --port 7331 .
codegenome rules --client all --dry-run .

Guides

• Cursor + MCP setup
• MCP server reference