Codebase Memory MCP is a local code-intelligence server that indexes software repositories into a searchable graph for AI coding agents.
The tool is built around the Model Context Protocol and gives agents structural queries over a codebase: architecture summaries, symbol search, code search, call tracing, impact analysis, index coverage, dead-code detection, and related graph operations. Instead of reading a project file by file, an agent can ask for likely entry points, hotspots, callers, callees, and neighboring tests, then verify those paths directly in source before editing.
Codebase Memory MCP is designed as a native executable with local indexing and optional agent-client integration. Its upstream installer can write agent configuration, run a coordination daemon, enable watching, and expose a local graph UI, so cautious adoption should distinguish the tool's code-map value from its global installation behavior.
I am testing Codebase Memory MCP as a bounded discovery layer, not as an authority layer. The value is faster orientation; the discipline is keeping policy, launch gates, and source reads above anything a graph tool says.