Codebase Memory MCP — Full Code Intelligence Engine
How It Works
Codebase Memory MCP parses your entire codebase using tree-sitter AST grammars and builds a persistent knowledge graph stored in SQLite. AI agents query this graph via 14 MCP tools to understand code structure, trace dependencies, and find patterns.
Key Features
- 66 language support — vendored tree-sitter grammars for JS, Python, Go, Rust, Java, C++, and more
- Sub-millisecond queries — SQLite-backed graph with Cypher-like query support
- 14 MCP tools — search, tracing, architecture analysis, dead code detection
- 3D graph visualization — built-in UI to explore code relationships
- Git diff impact mapping — understand how changes affect the codebase
- Cross-service HTTP route linking — maps API routes across microservices
- ADR management — Architecture Decision Records tracking
- Background file watcher — auto-sync when files change
- Single static binary — zero dependencies, just download and run
Performance Benchmarks
| Codebase | Files | Lines | Index Time |
|---|---|---|---|
| Average repo | ~500 | ~50K | < 1 second |
| Large monorepo | ~5,000 | ~500K | ~30 seconds |
| Linux kernel | 75,000 | 28M | ~3 minutes |
MCP Tools Available
- search — find functions, classes, variables by name or pattern
- trace — follow call chains and dependency paths
- architecture — get high-level module overview
- dead-code — detect unused functions and imports
- impact — analyze which code is affected by a change
- routes — map HTTP endpoints across services
- And 8 more specialized analysis tools
Installation
Download the single binary from GitHub Releases:
# Download latest release (check GitHub for your platform)
curl -LO https://github.com/DeusData/codebase-memory-mcp/releases/latest/download/codebase-memory-mcp-linux-amd64
chmod +x codebase-memory-mcp-linux-amd64
mv codebase-memory-mcp-linux-amd64 /usr/local/bin/codebase-memory-mcpFAQ
Q: What is Codebase Memory MCP? A: A high-performance MCP server that indexes your entire codebase into a queryable knowledge graph, giving AI agents deep code understanding with sub-millisecond response times.
Q: Is Codebase Memory MCP free? A: Yes, fully open-source under the MIT license.
Q: How many programming languages does it support? A: 66 languages via vendored tree-sitter grammars, covering all major languages including JavaScript, Python, Go, Rust, Java, C++, TypeScript, Ruby, and more.