🇬🇧 English · 🇨🇳 中文
"The map is not the territory." — Alfred Korzybski
Training data is the map. Your codebase is the territory. Sverklo gives the agent the territory.
The only code-intel MCP with a published benchmark and reproducible eval harness. Local-first MCP server that gives Claude Code, Cursor, Windsurf, and Zed a real symbol graph, a blast-radius lens, and a git-pinned memory — so the agent stops guessing. MIT. Zero config. Your code never leaves the machine.
Paper (Zenodo, CC BY 4.0) · bench:primitives — 62× fewer tokens than naive grep, 2.9× fewer than tuned grep, single tool call vs grep's 7-12 (n=60) · bench:swe — 38/65 perfect recall on 5 OSS repos, including the runs we lose.
One-click install
⭐ If sverklo saved your AI from hallucinating, please star this repo — it's the single most useful thing you can do to help others find it. Then share with one teammate who's tired of getUserByEmail() not existing in their codebase.
▶ Watch the 90-second demo on YouTube — sverklo init → audit → 37 MCP tools → live Claude Code integration with blast-radius and symbol-graph traversal.
Why
Your AI agent edits UserService.validate(). It doesn't know 47 other functions call it. It hallucinates getUserByEmail() because that's how its training data spelled it — your code uses findByEmail(). It forgets the design decision you made yesterday because context was compacted. Tests pass because they mock the dependency. Breaking changes ship.
Sverklo drills into your repo before the agent does — symbol graph, blast radius, semantic recall, and git-pinned memory — so the agent reasons about your code instead of pattern-matching from training data.
37
MCP tools your agent uses | < 1 s
incremental refresh after each edit | 0 bytes
of your code leave the machine |
npm install -g sverklo
cd your-project && sverklo init