SYKE
Open Source AI Onboarding GitHub
SYKE v1.8.11

Open-source guardrails
for AI coding
that runs locally.

SYKE gives Claude Code, Cursor, Windsurf, and other MCP clients a live dependency map, local impact analysis, and a build gate before changes spread through your codebase.

Without SYKE
AI: "Fixed auth.ts. Done!"
⚠ UserService.ts — import broken
⚠ SessionStore.ts — type mismatch
⚠ api/middleware.ts — undefined ref
+19 more files broken — discovered 3 days later
With SYKE
✓ auth.ts → 22 dependents detected
✓ gate_build: FAIL — 15 cascade errors blocked
✓ "3-day bug prevented in 3 seconds."
▶
Real-time Safety Gate
PASS / WARN / FAIL — your AI's last checkpoint before every build
◆
Zero-Config Analysis
8 languages auto-detected. Full dependency graph in seconds, no setup.
■
Visual Dependency Map
See every connection. Simulate file changes. Refactor with confidence.
◆
5-Min AI Onboarding
New hire joins Friday. They ship Monday. scan_project maps the full codebase in minutes.
8 languages
Zero config
MIT open source
One command to start
npx @syke1/mcp-server@latest
View Repository

5-minute setup. No subscription. Optional BYOK only for semantic AI workflows.

OPEN SOURCE

SYKE now ships as a fully open-source MIT project. Install it locally for free, bring your own AI key only if you want semantic analysis, and contribute through the GitHub repo.

View Repository

Works with your favorite AI coding tools

Claude Code Cursor Windsurf Copilot

🔒 Your source code stays on your machine unless you explicitly send an optional BYOK prompt to your own AI provider.

Click any node in the graph to see its impact analysis.

auth/auth_repository.dart
Layer BE
Risk HIGH
Deps 9 dependents
Imports 3
DESCRIPTION

IMPACT ANALYSIS
DIRECT (0)
    TRANSITIVE (0)
      GATE CHECK
      ⚠ Modifying this file would trigger WARN from gate_build (9 impacts)

      Why SYKE Exists

      THE PROBLEM
      • AI edits code blindly
        Claude Code, Cursor, and Copilot are powerful — but they have no awareness of your dependency graph. They modify files without knowing what breaks downstream.
      • Without SYKE: silent failures
        Broken imports. Cascade type errors. Build failures that only appear 10 commits later. The AI moved fast and broke things — in files it never even touched.
      THE SOLUTION
      • MCP Server — eyes for your AI
        SYKE plugs into your AI coding tool as an MCP Server. Before any file is changed, the AI calls check_safe and gets a full cascade map.
      • Dependency graph, always current
        Multi-language static analysis across 8 languages. Parses your entire codebase into a live dependency graph that the AI can query — not a snapshot, not a config file.
      • gate_build — mandatory build gate
        A hard PASS / WARN / FAIL verdict runs before every build. If impact is HIGH, the build is blocked. No more "it compiled but everything is broken."
      WHY DEVELOPERS CHOOSE SYKE
      • 60 seconds to full protection
        One command. Zero config files. SYKE auto-detects your language, builds the entire dependency graph, and starts protecting your codebase. Nothing to install, nothing to configure.
      • Your code never leaves your machine
        SYKE runs 100% locally. No cloud uploads, no telemetry, and no third-party access to your source code unless you explicitly use an optional BYOK AI workflow.
      • Plugs into your existing workflow
        Claude Code, Cursor, Windsurf, Copilot — SYKE works as an MCP server inside the tools you already use. No new IDE, no new CLI, no learning curve.
      YOUR CODE NEVER LEAVES YOUR MACHINE
      YOUR MACHINE
      💻 Your Source Code STAYS HERE
      ↓
      ⚙ SYKE MCP Server Static analysis • Dependency graph • Impact scoring
      ↓
      🎨 3D Dashboard localhost:3333
      Only optional BYOK traffic crosses the network ↓
      🔒 Selected code + provider prompt
      YOUR AI PROVIDER
      ✅ Optional Semantic Analysis runs only when you call ai_analyze or scan_project
      🛡
      Local by default. No telemetry. No background cloud sync. No hidden proxying.
      Your AST, dependency graph, and structural analysis stay on your machine unless you explicitly send selected context to your own AI provider.
      UNDER THE HOOD — 5 ALGORITHMS, ZERO AI TOKENS

      Most tools stop at counting imports. SYKE runs five graph algorithms locally — no API calls, no token cost, instant results.

      • SCC Condensation + Topological Sort
        Tarjan's algorithm detects every circular dependency cluster, condenses them into a clean DAG, then computes correct cascade levels. Before: "47 files affected" (inflated by cycles)
        After: "3 in cluster (L0) → 5 (L1) → 4 (L2)"
      • Composite Risk Score (0–1)
        Five weighted signals in one score: fan-in (30%), stability index (20%), cyclomatic complexity (20%), cascade depth (15%), and PageRank (15%). Your AI gets a calibrated number, not just "HIGH." auth_service → 0.82 CRITICAL | utils → 0.31 LOW
      • Git Change Coupling
        Mines your commit history to find files that always change together — even without import relationships. Catches hidden dependencies that static analysis misses entirely. config.json (85% co-change) | auth.css (72%)
      • PageRank for File Importance
        Google's PageRank adapted for code. A file imported by many important files ranks higher than one imported by many leaf files. Every file gets a rank and percentile. auth.ts: rank #1, 99th pctl | utils.ts: rank #4, 92nd
      • Incremental Updates + Memoized Queries
        One file changed? Only that file is re-parsed. BFS results are cached with smart invalidation. Large codebases stay fast — 50ms updates instead of 2-second rebuilds. 10K files: full rebuild 2s → incremental 50ms
      5-MINUTE ONBOARDING BYOK

      New team member joins your project.
      They're productive by end of day.

      Traditional onboarding burns 2–4 weeks of senior developer time walking through architecture, module ownership, and danger zones. scan_project generates that entire document in under 5 minutes — automatically.

      WITHOUT SCAN_PROJECT
      → Day 1–3: Architecture walkthrough with a senior dev
      → Week 1–2: Mapping modules, owners, conventions
      → Week 3–4: First cautious PR. Still asking questions.
      2–4 weeks lost
      WITH SCAN_PROJECT
      ✓ Minute 1: Run scan_project
      ✓ Minute 5: Full architecture doc — modules, risk zones, entry points
      ✓ Day 1: First meaningful PR shipped confidently
      5 minutes. Ship same day.
      • Full architecture document, auto-generated
        Directory structure, module ownership, high-risk files, circular dependencies, entry points, and a quick-start guide — written in plain English by AI, grounded in your live dependency graph. scan_project → ARCHITECTURE.md (auto-written)
      • Turns tribal knowledge into a repeatable document
        Use your own model key to regenerate architecture context whenever the codebase changes, instead of repeating the same onboarding walkthrough by hand.
      Learn more about AI onboarding →
      LOCAL GRAPH ANALYSIS
      • Real-time monitoring
        Catches cascade issues as the AI codes, not after the build fails. Every file save triggers an instant impact re-analysis.
      • Advanced graph algorithms
        PageRank hub ranking, composite risk scoring, SCC circular dependency detection, and cascade-level analysis on the condensed DAG.
      OPTIONAL BYOK AI
      • AI-powered semantic analysis
        Gemini, OpenAI, or Claude (BYOK) reads the impacted files and flags broken imports, type mismatches, and interface violations before you even run the compiler.
      • Full codebase onboarding
        scan_project generates a comprehensive architecture document — directory structure, module map, danger zones, and quick-start guide — so new team members ramp up in minutes, not days.

      Open Source + BYOK

      OPEN SOURCE 100% free
      • Local MCP server with dependency graph analysis
      • Structural tools such as gate_build, analyze_impact, get_hub_files, and check_warnings
      • Local dashboard and graph refresh workflow
      • 8-language support with zero-config detection

      No subscription, no license gate, no checkout flow.

      OPTIONAL AI BYOK
      • ai_analyze for semantic cascade analysis
      • scan_project for onboarding-document generation
      • Use Gemini, OpenAI, or Anthropic with your own key
      • Provider usage and limits depend on the model you choose
      See AI Onboarding

      SYKE does not resell model access or add token markup.

      COMMUNITY Support the project
      • Star the repository and file issues
      • Open pull requests and contribute fixes
      • Sponsor maintenance if you want to fund development
      • Share examples, docs, and real-world feedback
      Open GitHub Read FAQ

      Support happens in the open rather than through a subscription tier.

      OPEN ACCESS

      The public MCP workflow is now aligned around open source + 100% free local access + BYOK.
      Install it locally, bring your own AI key only if you want optional semantic analysis, and contribute through GitHub if you want to support ongoing maintenance.

      View Repository

      No subscription terms, purchase workflow, or feature-unlock payment is required for the local server.

      Feature Availability
      Feature Local Core Optional BYOK AI
      gate_build (PASS/WARN/FAIL) ✓ —
      check_safe ✓ —
      get_dependencies ✓ —
      analyze_impact + risk scoring ✓ —
      get_hub_files (PageRank) ✓ —
      check_warnings + refresh_graph ✓ —
      ai_analyze (Gemini/OpenAI/Claude) — ✓
      scan_project (codebase onboarding) — ✓

      Quick Start

      Step 1 — Optional AI config
      // Optional ~/.syke/config.json
{
  "geminiKey": "your-gemini-api-key"
}

      You can skip this entirely if you only want local structural analysis. Optional AI keys include geminiKey, openaiKey, and anthropicKey.

      Step 2 — Register MCP server
      # Claude Code
claude mcp add syke -- npx @syke1/mcp-server@latest
      Cursor / Windsurf / Other IDEs

      Cursor (.cursor/mcp.json):

      {
  "mcpServers": {
    "syke": {
      "command": "npx",
      "args": ["@syke1/mcp-server@latest"]
    }
  }
}

      Windsurf (~/.codeium/windsurf/mcp_config.json): same format as Cursor.

      Windows: If npx is not found, use: "command": "C:\\Program Files\\nodejs\\npx.cmd"

      Step 3 — Add build gate
      # Add to your project's CLAUDE.md
After completing code changes, always run the gate_build MCP tool before committing or deploying.

      Your AI will automatically run SYKE's safety check after every task — catches what AI missed before you ship.

      Step 4 — Restart & open dashboard
      Restart your AI agent. Open http://localhost:3333 to see your live dependency graph.
      Available Tools
      • gate_build
      • analyze_impact
      • check_safe
      • get_dependencies
      • get_hub_files
      • refresh_graph
      • ai_analyze
      • check_warnings
      Troubleshooting
      • Port 3333 already in use Set a different port via environment variable: 
        "SYKE_WEB_PORT": "3334"  // in your MCP config env
      • "Could not detect project root" Specify your project root explicitly: 
        SYKE_ROOT=/path/to/your/project npx @syke1/mcp-server@latest
      • Windows: native module build error Requires Node.js 18+ LTS. If you see build errors, try running in PowerShell (Admin): 
        npm install -g node-gyp
      Tool Reference
      Tool Description
      gate_build Mandatory PASS/WARN/FAIL check before builds
      analyze_impact Show all files affected by modifying a target
      check_safe Quick safety check for a single file
      get_dependencies List imports and dependents of a file
      get_hub_files Find the most connected files in your project
      refresh_graph Force re-scan of the dependency graph
      check_warnings Detect potential issues across the codebase
      ai_analyze (BYOK) AI-powered semantic analysis of cascade impacts using your own provider key
      scan_project (BYOK) Generate a full codebase onboarding document with your chosen AI provider
      87 nodes · 222 edges · 6 layers
      SYKE v1.8.11
      Scroll to zoom
      Drag to rotate
      Terms · Privacy · Security · Support · FAQ · Changelog · GitHub
      support@syke.cloud
      © 2026 SYKE. Released as open-source software.