Player not loading? Watch on YouTube
This comparison examines Codebase-Memory-MCP alongside Graphify, GitNexus and CodeGraph as code knowledge graph tools for an AI agent. The speaker describes Codebase-Memory-MCP as an MIT-licensed static C binary with compiled tree-sitter grammars for 158 languages. It stores its index in SQLite and provides 15 MCP tools, arbitrary Cypher queries and embedded semantic search without an API key.
The comparison favors different tools for different work. The speaker prefers Graphify for non-code documents and graph exports, GitNexus for structured multi-repository workflows, and CodeGraph for file watching and front-end callback tracing. Codebase-Memory-MCP gets the recommendation for cross-service connections, including HTTP call matching and framework routing. Its Git polling updates the index at intervals of 5 to 60 seconds.
Performance figures are reported claims rather than guarantees. The speaker cites five structural queries that used about 3,400 tokens through Codebase-Memory-MCP versus 412,000 through grep exploration. A separate preprint result across 31 repositories reports a more conservative 10x reduction. The Linux kernel indexing example takes three minutes on an M3 Pro.
The limitations matter for a coding assistant: large indexes can consume substantial RAM, queries execute sequentially, and hybrid LSP resolution covers only 10 languages. The other 148 use syntactic resolution. It also lacks PDF, image and video indexing.
