CodeGraph slashes Claude Code tool calls 94%
A local-first MCP server that builds a pre-indexed semantic knowledge graph of codebases to optimize AI agent exploration. By replacing expensive file scanning with structured symbol queries, it claims to reduce Claude Code tool calls by up to 94% and speed up exploration by 77% while keeping all data entirely on-device.
CodeGraph is a masterclass in local-first context engineering, proving that structured symbol relationships are more valuable for coding agents than fuzzy vector embeddings.
- –Uses tree-sitter to build a high-fidelity map of functions, classes, and call hierarchies without external API dependencies or data leakage.
- –Replaces slow, token-intensive "blind" searches like grep and glob with deterministic graph lookups, dramatically lowering the "exploration tax" for agents.
- –Recent refactoring to remove vector search functionality in favor of pure symbol-graphing signals a growing consensus: for code, structural logic beats semantic similarity.
- –Automatic integration with Claude Code via the Model Context Protocol (MCP) makes it a seamless productivity boost for CLI-based development.
- –The project is currently seeing a massive surge in community interest, trending with over 400 stars in a single day as developers look for ways to make Claude Code more cost-effective.
DISCOVERED
1h ago
2026-05-17
PUBLISHED
1h ago
2026-05-17
RELEVANCE