codebase-memory-mcp: Build a Searchable Code Knowledge Graph

Overview

codebase-memory-mcp is a code intelligence server for MCP-compatible coding agents. It parses a repository and stores relationships such as definitions, calls, imports, routes, and types in a persistent knowledge graph, giving agents a structured way to explore a codebase.

It is aimed at developers who want agents to navigate unfamiliar or large repositories without repeatedly reading and searching files. The server runs locally and supplies graph queries to the agent; it does not include its own conversational LLM.

Key Features

• Indexes source code using bundled tree-sitter grammars, with Hybrid LSP semantic type resolution for supported languages.
• Provides MCP tools for structural search, call tracing, architecture summaries, code search, change-impact analysis, and graph queries.
• Stores graph data locally in SQLite and supports automatic change detection and re-indexing.
• Offers semantic and full-text search alongside structural graph queries.
• Can link HTTP routes and calls across services, and index Dockerfiles and Kubernetes resources.
• Includes an optional browser-based graph visualization UI.
• Supports one-shot CLI access as well as integration with multiple coding-agent clients.

Use Cases

• A developer joining a large or unfamiliar project can ask an MCP-compatible agent to trace a function, find callers, or summarize architecture.
• A maintainer preparing a change can inspect affected symbols and call relationships before editing.
• Teams using coding agents can provide local, structured code context without sending repository contents to a hosted service.
• Engineers working across services can explore route-to-call relationships and cross-repository links.

Getting Started

On macOS or Linux, install with:

curl -fsSL https://raw.githubusercontent.com/DeusData/codebase-memory-mcp/main/install.sh | bash

The installer can configure detected agent clients. Restart your agent and ask it to index the project. See the README for Windows instructions, manual setup, and configuration options.

Alternatives

• codegraph: CodeGraph also builds a local code-structure graph for agent queries, with multi-language support and call-path and change-impact retrieval.
• GitNexus: GitNexus creates interactive repository graphs and offers a Graph RAG agent in the browser, rather than providing a local MCP server.
• Ferret MCP: Ferret MCP combines static analysis with AI interpretation to explain architecture and APIs, rather than centering on a persistent code-relationship graph.

Considerations

• This is a structural analysis backend, not a standalone chat assistant. Use it with an MCP-compatible client that provides the language-model interaction.
• Indexing and graph queries operate locally, so users need to install the executable and allow it to access the repository and relevant agent configuration files.
• Semantic type resolution is available for a subset of languages; parsing coverage and semantic detail can vary by language and project.
• The project exposes many tools and client integrations. Review its configuration and security documentation if you want to audit what the installer changes before running it.