agent-context-code

A 100% local semantic code search tool featuring hybrid search (vector + BM25), AST-aware chunking, and native MCP integration for AI coding assistants.

Local semantic code search for AI coding assistants.

Release status: 0.9.0 beta. The project is well-tested and actively used, but broader real-world usage may surface edge cases we have not encountered yet.

AGENT Context Local is an MCP server that gives your AI coding assistant semantic understanding of your codebase. It parses code into functions and classes using tree-sitter, then combines keyword matching with vector similarity so you can search by meaning — "where do we validate auth tokens?" — instead of relying on grep or burning context window limits on file-by-file exploration.

Everything runs completely on your local machine. No API keys, no telemetry, no cloud dependencies.

Source: tlines2016/agent-context-code

Key Features

• Hybrid search — Keyword matching and vector similarity combined for results that are both precise and semantically relevant.
• AST-aware chunking — Tree-sitter parses your code into functions, classes, and methods. No arbitrary line splits, no broken context.
• 100% local — Embeddings generated on-device, embedded vector database, zero API calls, zero uploads. Works perfectly in air-gapped, compliance-restricted, and proprietary IP environments.
• Incremental indexing — Content-hash tracking ensures only modified files get re-indexed. Re-indexing after a small change takes seconds.
• Graph-enriched results — Search results include structural context like class hierar