Cut AI token costs 95%+ on code exploration. The leading MCP server for precise, symbol-level GitHub code retrieval via tree-sitter AST. Works with Claude Code, Cursor & any MCP client. 313B+ tokens saved.

(ACL 2025 Main) A Comprehensive Benchmark for Code Information Retrieval.

State aware knowledge compression, ingestion, and hybrid retrieval engine. Zero dependencies. Sub-100ms queries.

Verified code context for agents

ContextAtlas — context infrastructure for AI coding agents: hybrid retrieval, project memory and retrieval observability via CLI, MCP server or embeddable library. Tree-sitter indexing, LanceDB vector search, FTS5 and token-aware context packing.

Local code search for AI agents: six fast, purpose-built tools that return ranked answers, not raw grep. Because maybe grep isn't all you need... 🍬

This repository contains the implementations of our experiments and our approach presented in the paper: CoNCRA: A Convolutional Neural Network Code Retrieval Approach

CoQuIR: A Comprehensive Benchmark for Code Quality-Aware Information Retrieval

🌈 Paper-implementations in Code Search (Baseline).

Deterministic method-first retrieval for AI coding agents.

SkeletonGraph is a zero-LLM structural index for AI coding agents. It uses tree-sitter, cross-file call graphs, and PageRank centrality to fetch the exact function your agent needs, saving tokens and turns. Includes an MCP server for Cursor, Claude Code, Copilot, and Windsurf.

Adaptive q-log BM25 for code retrieval under fixed generic tokenization

Semantic Language-Indexed Code Extraction with Backward Slicing for Repository-Scale Code Generation

Semantic code retrieval engine for AI coding agents, with hybrid search, AST-aware chunking, graph expansion, and token-aware context packing.

Structured code retrieval for AI agents — index once with tree-sitter, query symbols precisely via MCP. Cut code-reading token costs by up to 99%.

Agent-safe code retrieval for MCP coding agents: runtime-first search, exact reads, call graph context, and fresh repo evidence before edits.

A production-grade LLM context compression and retrieval engine built entirely on the Python standard library. It solves the single biggest bottleneck in LLM agent effectiveness: context window waste.

AST-aware chunking of code for contextual retrieval into the Unison brain.