CodeTrellis — Give AI Full Project Awareness

Scan your codebase, compress to ~1K tokens, inject into every AI prompt.

Works with: GitHub Copilot | Claude | Cursor | Windsurf | any MCP-compatible AI

The Problem

• 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

Quick Start

pip install codetrellis
codetrellis scan /path/to/project --optimal
codetrellis init . --ai   # sets up Copilot/Claude/Cursor integration

How It Works

┌─────────────────────────────────────────────────────────────────┐
│                         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      │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

Top Features

• 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_practices tool — 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

📋 Full feature list

Installation

pip install codetrellis

# Optional extras
pip install codetrellis[all]     # AST parsing, YAML, color, token counting
pip install codetrellis[ast]     # Tree-sitter AST parsing only

Output Tiers

# 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

CLI Commands

# 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 caches

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

MIT License — Keshav Chaudhary 2026