Codebase Memory MCP
A local code intelligence MCP server gives coding agents a persistent code graph and on-device semantic search on macOS, Linux and Windows.
Codebase Memory MCP maps a repository's structure so AI coding agents can explore dependencies without repeatedly reading whole files. It's for developers using Claude Code, Codex CLI, Cursor or other MCP-compatible clients.
- Code tracing: Follow calls and data flow across files and packages, inspect architecture, and connect HTTP routes with calls between services.
- Search and cleanup: Find code by meaning with on-device nomic-embed-code embeddings, locate near-duplicate functions, and detect functions with no callers while excluding recognized entry points.
- Change impact: Map uncommitted Git changes to affected symbols and assess how far those changes reach through the codebase.
- Cross-repository analysis: Link repositories in one store and inspect their combined services and dependencies, including Dockerfiles, Kubernetes manifests and Kustomize overlays.
- Shared context: Preserve architectural decisions alongside the graph, share graph snapshots with teammates, and explore relationships in an optional local 3D browser view.
The server is open source under MIT and runs on macOS, Linux and Windows without Docker, a language runtime, a hosted service or an API key. Parsing and semantic search happen locally. It has no embedded LLM; your chosen coding agent handles model access and any data it sends to its provider.
It parses languages including Python, TypeScript, Go, Java and Rust. Type-aware call resolution covers selected language families; others use textual resolution. A background watcher updates the index as files change.
