Sverklo MCP Server

Sverklo

Hallucination-resistant code intelligence for Claude Code, Cursor, Windsurf, and Zed. Semantic search, impact analysis, persistent memory — pick 3. Local, MIT, zero config.

Why

Your AI agent edits UserService.validate(). It doesn't know 47 other functions call it. It hallucinates an import. It forgets the design decision you made yesterday because context was compacted. Tests pass because they mock the dependency. Breaking changes ship.

Sverklo (Russian: сверкло, "drill") gives your agent a structural understanding of your codebase — symbol graph, blast radius, semantic recall, and git-pinned memory — so it stops guessing.

npm install -g sverklo
cd your-project && sverklo init

That's it. sverklo init auto-detects your installed AI coding agent (Claude Code, Cursor, Windsurf, Zed), writes the right MCP config, appends instructions to your CLAUDE.md, and runs sverklo doctor to verify the setup. No API keys. No cloud. Telemetry off by default.

First-run note. Sverklo's embedding model (all-MiniLM-L6-v2 ONNX, ~86 MB) is downloaded from HuggingFace on first use into ~/.sverklo/models/ and cached forever — every subsequent run is fully offline. Bundling the model into the npm tarball is on the v0.13 roadmap.

Want proof before installing? Browse the /report leaderboard — Sverklo audits of 47 popular OSS repos (express, react-hook-form, vite, lodash, prisma, …) with grade cards for dead code, circular deps, coupling, and security.

---

Grep vs Sverklo — the same question, side by side

Every one of these is a query a real engineer asked a real AI assistant last week. Grep gives you lines. Sverklo gives you a ranked answer.

If the answer to your question is "exact string X exists somewhere," grep wins. Use grep. If the answer is "which 5 files actually matter here, ranked by the graph," you need sverklo.

---

Works with every MCP editor

---

Hero tools

See all 23 tools below.

Search — find code without knowing the literal string

Impact — refactor without the regression

Review — diff-aware MR review with risk scoring

Memory — bi-temporal, git-aware, never stale

Index health

---

When to reach for sverklo — and when not to

We're honest about this. Sverklo isn't a magic 5x speedup and it doesn't replace grep. It's a sharper tool for specific jobs.

When sverklo earns its keep:

• You don't know exactly what to search for
• You need to prove dead code (zero references across the whole symbol graph)
• You need the blast radius of a refactor before you start
• You're reviewing a large PR and need to know what to read first

When grep is still the right tool:

• Exact string matching — "does this literal string exist?"
• Small codebases under ~50 source files — just read everything
• Single-file diffs — git diff + Read is hard to beat
• Build and test verification — only Bash runs npm test

If a launch post tells you a tool is great for everything, close the tab.

---

How It Works

graph LR
    A[Your Code] --> B[Parse<br/>10 languages]
    B --> C[Embed<br/>ONNX/Ollama]
    B --> D[Build Graph<br/>imports/exports]
    D --> E[PageRank<br/>importance]
    
    F[Agent Query] --> G[BM25]
    F --> H[Vector Search]
    E --> I[PageRank Boost]
    G --> J[RRF Fusion]
    H --> J
    I --> J
    J --> K[Token-Budgeted<br/>Response]

1. Parses your codebase into functions, classes, types (TS, JS, Python, Go, Rust, Java, C, C++, Ruby, PHP)
2. Embeds code using all-MiniLM-L6-v2 ONNX model (384d, fully local) — or any Ollama model via config
3. Builds a dependency graph and computes PageRank (structurally important files rank higher)
4. Searches using hybrid BM25 + vector similarity + PageRank, fused via Reciprocal Rank Fusion
5. Remembers decisions and patterns across sessions, linked to git state
6. Watches for file changes and updates incrementally

---

Performance

Real measurements on real codebases. Reproducible via npm run bench (methodology).

• Search p95 stays under 26 ms even on a 4k-file monorepo
• Impact analysis is sub-millisecond — indexed SQL join, not a string scan
• 10 languages: TS, JS, Python, Go, Rust, Java, C, C++, Ruby, PHP

---

Quick Start

Claude Code

npm install -g sverklo
cd your-project && sverklo init

Creates .mcp.json at your project root and appends sverklo instructions to CLAUDE.md. Safe to re-run. If sverklo doesn't appear in /mcp after restart, run sverklo doctor.

Cursor / Windsurf / VS Code / JetBrains

Use the full binary path (which sverklo) to avoid PATH issues in spawned subprocesses:

{
  "mcpServers": {
    "sverklo": {
      "command": "/full/path/to/sverklo",
      "args": ["."]
    }
  }
}

Config locations: .cursor/mcp.json, ~/.windsurf/mcp.json, .vscode/mcp.json, or JetBrains Settings -> Tools -> MCP Servers.

Antigravity

sverklo init writes the global config at ~/.gemini/antigravity/mcp_config.json. Because Antigravity lacks per-project MCP config, you'll need to re-run sverklo init from each project or use separate keys (sverklo-projA, sverklo-projB).

Any MCP client

npx sverklo /path/to/your/project

First run: The ONNX embedding model (~90 MB) downloads automatically. Takes ~30 seconds on first launch, then instant.

---

Why not... (as of 2026-04)

---

Audit formats

sverklo audit generates codebase health reports in six formats: markdown, html, json, sarif, csv, and badges. Run sverklo audit --format html --open for a self-contained report with god nodes, hub files, orphan detection, coupling analysis, and language distribution. Use sverklo audit --badge to add an A-F health grade shield to your README.

---

CLI tools

Sverklo ships a CLI for CI and local use: sverklo review --ci --fail-on high for risk-scored diff review (auto-detects PR ref in GitHub Actions), sverklo audit for codebase health reports, and a GitHub Action that posts review comments on PRs. Run sverklo audit-prompt or sverklo review-prompt to get battle-tested workflow prompts you can paste into any agent.

---

Telemetry

Off by default. Sverklo makes zero network calls unless you explicitly run sverklo telemetry enable. If you opt in, we collect only anonymous usage metrics (no code, no queries, no file paths). Full schema and implementation details in TELEMETRY.md.

---

Open Source, Open Core

The full MCP server is free and open source (MIT). All 23 tools, no limits, no telemetry, no "free tier" — that's not where the line is.

Sverklo Pro (later this year) adds smart auto-capture of decisions, cross-project pattern learning, and larger embedding models. Sverklo Team adds shared team memory and on-prem deployment.

The open-core line: Pro adds new things, never gates current things. Anything in the OSS server today stays in the OSS server forever.

---

Links

• Website
• npm
• Issues
• First Run Guide
• Benchmarks

License

MIT