Srclight – Deep code indexing MCP server (FTS5 and Tree-sitter)
Tree-sitter + FTS5 + MCP = tokens saved for AI agents to actually code, not search.
Structural, byte-precise, token-cheap codebase access for coding agents — tree-sitter over a CLI and an MCP server.
WASM grammar loading without recompilation beats static tree-sitter bindings.
Developers building AI coding agents, teams using MCP-enabled tools
Cursor · Continue · Sourcegraph Cody
Tree-sitter + FTS5 + MCP = tokens saved for AI agents to actually code, not search.
Tree-sitter dependency graph saves 5,000-20,000 tokens per agent query vs exploration.
95% fewer tokens than grep for PyTorch retrieval—agent-first code search with proven benchmarks.
Pure Rust CPU-only code search with persistent index beats transformer-heavy alternatives.
Tree-sitter MCP cuts Claude code task costs 17–82% while improving accuracy.
MCP server gives Claude fast, token-efficient codebase knowledge without loading raw files.