Athena: local-first MCP server cuts agent context usage by 49x-2067x
Significant-Log-4272 · reddit · 2026-11-22
A Redditor released Athena v1.3.0 (Apache-2.0), a fully offline MCP server + CLI that gives coding agents a 2KB repo map, paged evidence bundles, and AST-level breaking-change checks across 14 languages.
Highlights:
- Context-safe paging: heavy tools return summaries; full bundles stored locally under content-addressed IDs for on-demand inspection.
- 2KB code graph: incremental local graph with trigram and hub-guarded BFS queries; dynamic calls like getattr/eval marked as explicit unknowns.
- Blast-radius checks: compares old vs new signatures, flags removed exports, shifted params, unguarded None defaults, recursion bugs, and traces affected callers.
- Proper MCP annotations: 12 tools with readOnlyHint/destructiveHint/idempotentHint/openWorldHint plus discovertools progressive discovery.
- One-command setup: init-mcp for Claude Desktop/Cursor/VS Code; transactional skill/hook installs for Claude Code, Codex, and others.
Tested offline on 10 repos with 10k+ stars (1,000 commits, 8,804 files): 0 errors, agent context tokens reduced 49x–2067x.
More from coding & agent
- Matt Pocock floats a 'perfect TypeScript codebase' agent skill and asks what belongs in it — mattpocockuk · 2026-11-22
- Dystalgia: a ComfyUI-based photo editor with context-aware cropping — aurelm · 2026-11-22
- Esri open-sources CityEngine C++ SDK for procedural 3D building generation — Promptmethus · 2026-11-22
- Influzer adds MCP allowlist export: mergeable policy packs from Top 100 servers — Fine_Airline_6832 · 2026-11-22
- Google testing Inbox mode for Antigravity to shift users into approving AI outcomes — testingcatalog · 2026-11-22
- Anthropic publishes Zero Trust security framework for enterprise AI agents — matthew_d_green · 2026-11-22