archex benchmarks its MCP code-retrieval server against cocoindex-code and Graphify
tom_mathews · reddit · 2026-07-23
archex is a local, deterministic retrieval and context-assembly engine for coding agents. It parses code with tree-sitter, chunks and embeds it, expands a dependency graph, and returns a token-budgeted bundle with a receipt showing what was included, what was skipped, and whether the bundle is safe to act on.
The author benchmarked it against cocoindex-code and Graphify on 19 external-repo tasks:
- Required-file recall: 0.95 / 0.32 / 0.70
- Missed-task rate: 0.16 / 0.79 / n/a
- Token efficiency: 0.76 / 0.48 / n/a
- Completion-penalty tokens: 922 / 11,188 / n/a
- Cold-start: 0 ms / 4.7 s / 937 ms
It ships as an MCP server with 17 tools, plus CLI, Python API, Docker support, 26 language support, no hosted inference in the core path, Apache 2.0 licensing, and 3,619 tests. The repo includes a reproducible benchmark command and a full methodology doc.
Related event: Archex Releases Local Code Context MCP Server for Coding Agents(3 posts)→
More from coding & agent
- Cognition's SWE-2 uses a KKT duality argument in RL to shift the effort Pareto curve — YouJiacheng · 2026-09-11
- First-ever Three.js Conference lands in Paris, with a panel on AI-shortened design workflows — OdinLovis · 2026-09-11
- Data engineering, not agent frameworks, is the real bottleneck for enterprise AI agents — dhruv2038 · 2026-09-11
- RTK Terminal Compression Cuts Tokens but Leaves Your AI Coding Bill Unchanged — Bartaseth · 2026-09-11
- GPT-6 Astra beats Factorio with enemies in 44 in-game hours at ~$4,500 API cost — liminal_bardo · 2026-09-11
- Investment Analyst Asks How to Build a Claude-Based Diligence Agent Stack — Careless_Tie2286 · 2026-09-11