AI library

Source · repo

codebase-memory-mcp: a local code knowledge graph for agents

Author
DeusData
Published
16 October 2026
Added
30 October 2026
Original
https://github.com/DeusData/codebase-memory-mcp

Summary

codebase-memory-mcp is an MIT-licensed MCP server that indexes a repository's code, docs, and infrastructure files into a persistent graph on your machine, so an agent can trace calls, dependencies, and the impact of a change without re-reading files. It needs no API key and runs locally. The latest release, v0.12.1 (2026-10-16), fixed installs and added Helm charts to the graph.

Key points

  • Install: run the installer (curl -fsSL https://raw.githubusercontent.com/DeusData/codebase-memory-mcp/main/install.sh | bash, or install.ps1 on Windows), restart Claude Code, and ask it to index the project. It is also on npm, PyPI, Homebrew, and go install. Manual setup: add an mcpServers entry pointing at the binary in ~/.claude.json or the project's .mcp.json, then check /mcp.
  • Tools: 17 in all, including index_repository, search_graph, trace_path (call paths), detect_changes (what a diff affects), get_architecture, and manage_adr (architecture decision records).
  • What it is good for: "what calls this?" and "what breaks if I change it?" questions in large or multi-language repositories, and infrastructure graphs (for example HelmRelease to Chart to Deployment to Image).
  • Optional: CBM Lens, a browser visualiser in open beta (v0.0.1).
  • Caveats: the binaries are large (over 300 MB in v0.12.0), and the first index after an upgrade is a full rebuild. Read the install script before piping it to a shell.
  • Around 47,800 stars (+550 in the week to 2026-10-30) and frequent releases.

Why it matters

It is a structural alternative to grep-style exploration: the agent asks a graph instead of spending context reading files, which helps most in large codebases.