contextmesh-cli 0.1.0

Change Data Capture for AI-assisted codebases. Keeps context files alive as your code evolves.

🧠 ContextMesh

The Change Data Capture (CDC) engine for AI Context. Stop feeding your agents stale docs.

---

If your AI coding agents (Claude Code, Cursor, Devin) are hallucinating import paths or struggling with multi-file refactors, it's because static context files rot.

ContextMesh is an active, ast-aware framework that monitors your codebase and surgically patches CONTEXT.md files as you code, ensuring your AI always has 100% accurate, hierarchical understanding of your repository.

It is not a scraper. It is a real-time CDC engine.

---

🚀 The 10x Claim (Benchmarks vs Static Scrapers)

We tested ContextMesh against monolithic generation approaches (e.g., Graphifyy) on a massive Django codebase:

1. 💸 99% Cheaper per Commit: Because ContextMesh uses surgical AST Diff-patching (--from HEAD~1), updating a core feature cost 4,275 tokens ($0.00). Re-running a traditional scraper on a large repo takes ~2,000,000 tokens.
2. ⚡ Native IDE Speeds: Full scaffold takes several minutes. Evaluating a local save takes ~1.2 seconds.
3. 🛡️ 100% Elimination of Context Drift: Our contextmesh validate engine parses the entire hierarchy. During testing, it caught 34 semantic drift issues (missing children directories, broken lateral links) that markdown scrapers missed entirely.

---

🔥 Features

• Ghost Watcher Daemon: Run contextmesh watch (or use the VSCode extension) and it will debounce filesystem events, automatically patching context files in the background without user intervention.
• Hierarchical Context Tree: Builds a parent-child mesh of CONTEXT.md files so your AI understands local logic and horizontal dependencies (e.g. auth depends on database).
• VSCode Extension & MCP Server: Don't like CLI tools? Use the VSCode Extension! It natively bridges to the Engine via the Model Context Protocol (MCP), giving Cursor/Claude Code native access to Health Scores and Auto-Fix capabilities.

🛠️ Usage

CLI Installation

pipx install contextmesh-cli
contextmesh init

1. Scaffold your Project

# Generate the semantic mesh for the first time
contextmesh scaffold .

2. Auto-Maintain (The Ghost Watcher)

# Run this, and never worry about context decay again.
contextmesh watch

3. Native IDE Integration (MCP)

ContextMesh acts as a standard MCP server. You can allow Cursor/Claude to natively invoke health checks and context regeneration via STDIO:

{
  "mcpServers": {
    "contextmesh": {
      "command": "contextmesh",
      "args": ["mcp-serve"]
    }
  }
}

🧠 Architecture

1. Diff Analyzer: Catches filesystem or git changes.
2. Salience Classifier: Rejects trivial changes (e.g., changing a string). Only passes structural AST changes.
3. LLM Patcher: Surgically modifies exactly the portion of the CONTEXT.md that changed.
4. Cross-Doc Validator: Runs rules over the tree to ensure parents don't hallucinate missing children.

🤝 Next.js / Cloud Dashboard

Coming soon in Phase 4 for enterprise governance and tracking team context-health.