Codebase Memory MCP is a local code-intelligence server that converts software repositories into persistent knowledge graphs for AI coding agents. It indexes functions, classes, call relationships, HTTP routes, infrastructure definitions, and cross-service links across many programming languages, allowing agents to explore architecture without repeatedly reading large numbers of files.

The project uses Tree-sitter-based parsing and exposes its graph through MCP tools for search, call tracing, architecture analysis, impact analysis, dead-code detection, Cypher queries, ADR management, and other structural tasks. It is designed for fast local indexing, supports a large set of language grammars, and includes a built-in 3D graph visualization interface. The project emphasizes that source code remains on the user’s machine and does not require hosted APIs or model keys.

Codebase Memory MCP is intended for coding agents working with medium or large repositories where file-by-file exploration consumes substantial context and tokens. By giving agents a compact structural representation of a codebase, it can improve navigation, dependency reasoning, and change analysis while remaining compatible with many MCP-capable development environments.