tesla-skill

Connect your Tesla to any AI agent. Voice or chat, any client, official Tesla API + Virtual Key signing.

You: "What's my Tesla's battery?"
AI:  Calls get_car_status → "78%, range 409 km. Climate is off, doors locked."

You: "Open climate to 22 degrees"
AI:  Calls set_climate(on=true, temp_c=22) → "Started, target 22°C."

You: "Honk so I can find it"
AI:  Calls honk_horn → "Honked. The car is to your right."

Why wearables + Tesla

Your agent goes wherever you go — smart glasses, smartwatch, AI pendant, earbuds. All of them speak MCP. None of them want you fumbling for your phone. With tesla-skill, controlling your car is one sentence away from whatever's on your wrist or collar:

• 🛒 Both hands on a supermarket cart: "Open the trunk."
• 🌧️ Kid on one arm, umbrella on the other: "Unlock the doors, AC to 22°C."
• ❄️ Office at 4:50pm in January: "Preheat the car."
• 🅿️ Stadium parking after a concert: "Flash the lights so I can spot it."
• 💬 Mid-conversation, can't grab your phone: "How much range have I got left?"

Your phone is busy / wet / dead / left in another room. Your wearable isn't.

What it is

A Model Context Protocol (MCP) server that exposes 10 tools any MCP-compatible AI agent can use to query and control your Tesla:

🔍 Read	🎛️ Control
get_car_status — battery, range, climate, lock, sentry	set_climate(on, temp_c)
get_car_location — lat/lng/speed/heading	lock_car / unlock_car
get_charge_info — SOC, limit, kW, ETA	start_charging / stop_charging
	flash_lights / honk_horn

Works with Claude Desktop, Claude Code, Cursor, Codex, Continue.dev, and any other MCP-compatible client.

Install with a coding agent

Send your agent (Claude Code, Codex, …) this repo URL and say "install this":

https://github.com/Vibetool/tesla-skill

The agent reads SKILL.md and walks you through setup — about 30 minutes of work (plus 1-3 days for Tesla developer review).

Or follow docs/SETUP.md yourself.

What you'll need

• A Tesla (2021+ for full Fleet API support)
• A Tesla owner account (the one in your Tesla mobile app)
• A public domain you control — Tesla requires it for hosting your Virtual Key public key. Free static hosts work (Cloudflare Pages, GitHub Pages w/ custom domain).
• Python 3.11+ on the machine running the MCP server
• Go 1.22+ only if you want control commands (read-only works without)

Architecture

       Your AI Agent
   (Claude / Cursor / etc.)
            │
            │  MCP over stdio
            ▼
   ┌──────────────────────────┐
   │   tesla_skill.server     │     10 tools (FastMCP)
   └──────────────────────────┘
            │
            ├──► Tesla Fleet API (REST)         → reads
            │
            └──► tesla-control subprocess        → control commands
                 (ECDSA P-256 signing,
                  Tesla's official Go CLI)

• Read commands call /api/1/vehicles/{id}/vehicle_data directly via httpx, with 30-second response caching to stay under Tesla's daily rate limit.
• Control commands shell out to tesla-control, which signs each command locally with your Virtual Key and POSTs to Tesla's /command/ endpoint. Sleeping cars are auto-woken with one retry.

Why this exists

Tesla deprecated the unsigned owner-API in 2024 in favor of Fleet API + TVCP signing. Most third-party Tesla integrations broke. This project gives you a clean MCP-shaped re-implementation:

• No middlemen — direct Tesla API, your tokens stay on your machine
• End-to-end encryption of stored tokens (Fernet at rest, OAuth in flight)
• Tesla-official signing path — wraps the same Go CLI Tesla recommends, no DIY crypto
• Mock mode for development without a real car
• Region-aware — China and global Fleet APIs both supported

See ABOUT.md for the design rationale.

Quick start (development, no Tesla account needed)

git clone https://github.com/Vibetool/tesla-skill.git
cd tesla-skill
python3 -m venv .venv && source .venv/bin/activate
pip install -e .

# Mock mode — fake car responds to all tools
echo "USE_MOCK=true" > .env
python -m tesla_skill.server  # stdio MCP server

# Or with the official MCP Inspector UI
npx @modelcontextprotocol/inspector python -m tesla_skill.server

For real-Tesla setup, see SKILL.md (agent walkthrough) or docs/SETUP.md (manual).

Documentation

• SKILL.md — agent installation walkthrough (auto-followed when you say "install this" to your AI)
• docs/SETUP.md — manual setup guide for users without an agent
• docs/AGENTS.md — connecting from Claude Desktop / Cursor / Continue / Codex / etc.
• docs/DEPLOY.md — optional public server deployment

Project status

v0.1.0. The 10 tools listed above are implemented and tested against a real 2022 Model Y. Roadmap: data dashboards, multi-vehicle support, scheduling tools (preheat at 8am, etc.).

Contributing

PRs welcome. The whole thing is ~1000 lines of Python; should be approachable.

License

MIT — see LICENSE.

Disclaimers

Not affiliated with Tesla, Inc. Use at your own risk — control commands physically affect your vehicle. Always double-check with your agent before issuing irreversible commands (unlocking, disabling sentry, summoning).