codebone-mcp
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@codebone-mcpWhat does the billing domain look like? Show its routes and models."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
🦴 codebone
Real-time codebase knowledge graph for your macOS menu bar.
Stop dumping dozens of files into your prompt. Let the dog sniff it. 🐾
📉 Why codebone?
AI assistants typically read 10–20 files just to understand your architecture — burning thousands of tokens on boilerplate before writing a single line of code.
codebone runs in your macOS menu bar, watches every file save, and delivers a live architectural snapshot directly to Claude, Cursor, or Gemini via MCP.
Related MCP server: mcp-code-indexer
🚀 Download & Install
⬇️ Download
Drag codebone.app to /Applications — done. A 🦴 appears in your menu bar.
Click it → Select Project Folder... to start your first scan.
💾 ~650 MB total (~390 MB model + ~260 MB venv). Requires macOS 13+ on Apple Silicon.
Homebrew:
brew tap palusc/codebone
brew install codeboneBuild from source:
git clone https://github.com/palusc/codebone.git
cd codebone
./install.sh🐾 How It Works
Sniff — Watches your repo on every
Cmd + Swith battery-aware debouncing.Think — Apple Silicon Metal GPU extracts domains, models, routes, and events in
< 1s.Map — Builds a Semantic System Graph linking files by shared business logic — not just imports.
Serve — Streams precise architectural context to your AI via MCP or curl.
💬 Usage with AI Assistants
Prompt naturally:
"Implement authentication middleware for billing routes. Use codebone."
The model calls codebone_context() and receives the exact schemas, routes, and relationships it needs — no file dumping required.
codebone auto-registers during ./install.sh. For manual setup, add to claude_desktop_config.json or .cursor/mcp.json:
{
"mcpServers": {
"codebone": {
"command": "npx",
"args": ["-y", "codebone-mcp"]
}
}
}That's it — no Python paths, no manual configuration.
{
"mcpServers": {
"codebone": {
"command": "/Users/YOUR_USERNAME/Library/Application Support/codebone/venv/bin/python3",
"args": ["-m", "codebone_mcp.server"],
"env": { "CODEBONE_PORT": "8053" }
}
}
}Replace YOUR_USERNAME with the output of whoami.
# codebone — Codebase Architecture
Project: /Users/you/my_project | Files: 142 | Updated: 2026-09-23
## Business Domains
- Authentication & Identity (12 files: auth/router.py, auth/jwt.py, models/user.py …)
- Payment & Billing (8 files: billing/stripe.py, billing/webhook.py, models/invoice.py …)
- Order Fulfillment (15 files: orders/service.py, events/order_created.py …)
## Focused query: codebone_context(domain="billing")
→ Returns only billing files, models (Invoice, Subscription),
routes (POST /checkout/session), and events (InvoicePaid, PaymentFailed).# Full architectural context
curl http://localhost:8053/codebone/context
# Filter by domain, file, or entity
curl 'http://localhost:8053/codebone/context?domain=billing'
curl 'http://localhost:8053/codebone/context?file=payments'
curl 'http://localhost:8053/codebone/context?query=InvoiceCreated'🗺️ Live Graph UI
open http://localhost:8053/codebone/graph/uiInteractive node-link diagram of your codebase. Dark glassmorphism theme, real-time search (/), click-to-inspect drawer, and live sync on every file save.
🧠 Brain Options
Provider | Description |
Built-in (Qwen 0.5B) (default) | Metal GPU-accelerated. Zero cost, zero cloud. |
Local URL | Ollama, LM Studio, or any OpenAI-compatible endpoint. |
Cloud BYOK | Your own API key — OpenAI ( |
Custom .gguf | Any local GGUF model via file dialog (e.g. Qwen 7B, Llama 3). |
🦴 ➔ Settings ➔ Model to switch at any time.
🔄 Smart Scan Adoption
Renamed, moved, or branched your project? Never re-scan from scratch.
SHA-256 fingerprinting reuses unchanged files instantly. Only modified files are re-sniffed.
🦴 ➔ Adopt / Link Existing Scan...
🔒 Privacy & Security
100% Localhost — Binds to
127.0.0.1:8053only. Zero cloud, zero telemetry.Prompt Injection Defense — Untrusted code is isolated with escaped XML wrappers and strict anti-jailbreak directives.
Symlink Jail — Symlinks cannot escape the project root into sensitive system paths.
Full Disk Access Helper — Pre-flight TCC check with 1-click System Settings shortcut.
Battery-Aware — 0.5s debounce on AC, 15s on battery.
🗑️ Uninstall
Option A — Uninstaller App (recommended):
Double-click Uninstall codebone.app → Clean All Data. Removes all processes, databases, models, logs, and MCP entries. Then drag codebone.app to the Trash.
Option B — Terminal:
./uninstall.sh # full wipe
./uninstall.sh --keep-data # keep databases & model for fast reinstallReleased under the MIT License.
This server cannot be deployed
Maintenance
Related MCP Connectors
Give your AI agent a persistent map of your project's structure, dependencies, and bugs.
Codebase graphs, caller impact analysis, and recorded project context for AI coding agents.
Codebase intelligence for AI agents — dead code, blast radius, ownership.
Shared memory for coding agents. Stop re-explaining your codebase every session.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceEnables AI coding tools to query your live codebase for routes, import graph, domain context, and blast radius, eliminating hallucinations about project structure.87 npm78MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to intelligently navigate and understand codebases by providing instant file descriptions, semantic search, and context-aware recommendations, eliminating the need to repeatedly scan files.16 PyPI20MIT
- AlicenseNot gradedqualityBmaintenanceIndexes codebases and lets AI agents retrieve precise code snippets (functions, classes, routes) instead of reading entire files, reducing token usage and improving accuracy.166 npm7MIT
- AlicenseNot gradedqualityBmaintenanceProvides AI coding assistants with deep, semantic understanding of local codebases via AST-aware chunking, cross-repo symbol graphs, and architectural memory, enabling context-aware code search and dependency tracing.11MIT