Codebase Memory MCP is an MIT-licensed Model Context Protocol server from DeusData that indexes a repository into a persistent code knowledge graph for AI coding agents. Its public README describes tree-sitter parsing across 158 languages, Hybrid LSP semantic resolution for selected languages, and MCP tools for structural questions about functions, classes, call chains, routes, and project architecture. The value proposition is direct: instead of making an agent repeatedly grep, reread files, or pack large source excerpts into context, the server gives the agent a queryable map of the codebase.
The strongest fit is large or unfamiliar codebases where Claude Code, Cursor, Codex-style agents, or other MCP-aware coding environments need structure before editing. Codebase Memory MCP ships as a local static binary and is positioned around code intelligence, architecture overview, impact analysis, and lower-token exploration. The public repo and docs include high-traction source signals plus benchmark-style claims about indexing speed, query latency, and token reduction; buyers should treat those as source-reported claims to verify on their own projects, not universal performance guarantees.
Teams should evaluate the security and workflow boundary carefully before adoption. The tool reads local repositories, and its own documentation warns that it can write agent configuration files, so sensitive codebases need source review, permission review, and clear rules about which projects may be indexed. It also overlaps with memory, code-search, and knowledge-graph tools already tracked in the aicoolies ecosystem, so this page frames it specifically as a local codebase-structure MCP layer rather than a general vector database, hosted documentation product, or autonomous coding agent.