Scan your codebase, compress to ~1K tokens, inject into every AI prompt.
Works with: GitHub Copilot | Claude | Cursor | Windsurf | any MCP-compatible AI
- AI assistants read files one at a time — they never see your full project
- They don't know about existing components, schemas, or patterns
- You explain your project structure repeatedly in every conversation
- AI lacks business domain understanding — doesn't know why code exists
pip install codetrellis
codetrellis scan /path/to/project --optimal
codetrellis init . --ai # sets up Copilot/Claude/Cursor integration┌─────────────────────────────────────────────────────────────────┐
│ CodeTrellis WORKFLOW │
├─────────────────────────────────────────────────────────────────┤
│ │
│ 1. SCAN 2. COMPRESS 3. INJECT │
│ ───────── ─────────── ──────── │
│ │
│ Read every → Convert to → Add to every │
│ file in minimal AI prompt │
│ project tokens │
│ │
│ 187 lines → 30 tokens → Full awareness │
│ │
└─────────────────────────────────────────────────────────────────┘
- 120+ language/framework parsers — Python, TypeScript, Go, Rust, Java, C#, and more
- MCP server for real-time AI context injection (JSON-RPC 2.0)
- MCP
get_best_practicestool — returns framework-specific coding practices on demand (106 YAML practice files covering 25 ecosystems) - JIT context engine — delivers only relevant sections for the file you're editing
- Incremental builds — only re-extract changed files
- Best Practices Library — 4,500+ practices auto-selected for your stack
- Output tiers — from ~800 tokens (compact) to full code context (logic)
- CI/CD mode — deterministic, parallel builds for pipelines
- AI integration — auto-generates Copilot, Claude, Cursor, Windsurf configs
pip install codetrellis
# Optional extras
pip install codetrellis[all] # AST parsing, YAML, color, token counting
pip install codetrellis[ast] # Tree-sitter AST parsing only| Tier | Truncation | Tokens | Use Case |
|---|---|---|---|
compact |
Yes | ~800-2000 | Quick overview |
prompt |
NO | ~8000-15000 | Default AI injection (includes code logic!) |
full |
NO | ~15000+ | Detailed analysis |
logic |
NO | ~30000+ | Full code context |
json |
NO | Variable | Machine processing |
# Use tiers
codetrellis scan ./project --tier compact # Minimal
codetrellis scan ./project --tier prompt # Default (recommended)
codetrellis scan ./project --tier full # Everything
codetrellis scan ./project --tier logic # With function bodies# Scanning
codetrellis scan [path] # Scan project
codetrellis scan [path] --optimal # Maximum quality (recommended)
codetrellis scan [path] --incremental # Only changed files
codetrellis scan [path] --ci # CI/CD mode (deterministic + parallel)
codetrellis scan --remote <url> # Scan a remote git repo
# AI Integration
codetrellis init . --ai # Generate Copilot/Claude/Cursor configs
codetrellis init . --update-ai # Regenerate AI files (no re-scan)
codetrellis mcp --stdio # Start MCP server
codetrellis context path/to/file.py # JIT context for a file
codetrellis skills # Generate AI-executable skills
# View & Export
codetrellis show # Show full matrix
codetrellis prompt # Print prompt-ready matrix
codetrellis export --json # Export as JSON
# Quality & Maintenance
codetrellis verify [path] # Build quality gate
codetrellis validate [path] # Validate extraction completeness
codetrellis coverage [path] # Show extraction coverage
codetrellis watch # Auto-sync on file changes
codetrellis clean [path] # Clean cachesSee CONTRIBUTING.md for development setup and guidelines.
MIT License — Keshav Chaudhary 2026