TeslaMate MCP Server

A Model Context Protocol server that exposes your TeslaMate PostgreSQL database to MCP-aware AI clients (Claude Desktop, Cursor, etc.) over either stdio or streamable HTTP.

Features

• 20 tools — 18 predefined analytics queries (battery, charging, driving, efficiency, locations) plus run_sql and get_database_schema
• 6 prompts — one-click workflows for battery health, driving summary, charging behaviour, anomaly hunting, weather efficiency, and a quick status report
• 2 resources — teslamate://queries and teslamate://queries/{name} for catalog browsing without invoking a tool
• Hardened run_sql — runs inside a PostgreSQL READ ONLY transaction with statement_timeout, lock_timeout, and an automatic row cap
• Live schema introspection — get_database_schema reads information_schema at runtime; no stale JSON checked into the repo
• Two transports, one binary — teslamate-mcp stdio for local clients, teslamate-mcp http for remote
• Bearer-token auth with timing-safe comparison; /health probe for liveness checks
• Decimal → float JSON serialization so language models see numbers, not strings

Requirements

• TeslaMate already running against PostgreSQL
• Python 3.11+ for a local install, or Docker for a remote deployment

Install

git clone https://github.com/cobanov/teslamate-mcp.git
cd teslamate-mcp
cp env.example .env
# Edit .env — at minimum, set DATABASE_URL
uv sync

CLI

The teslamate-mcp console script has four subcommands:

teslamate-mcp stdio                          # local (Cursor / Claude Desktop)
teslamate-mcp http [--host] [--port]         # remote (HTTP / SSE)
teslamate-mcp gen-token                      # produce an AUTH_TOKEN value
teslamate-mcp list-tools                     # diagnostic: list registered tools

python -m teslamate_mcp <subcommand> works too.

Local use (stdio)

Configure your MCP client to launch the stdio server. Example for Cursor or Claude Desktop:

{
  "mcpServers": {
    "teslamate": {
      "command": "uv",
      "args": ["--directory", "/path/to/teslamate-mcp", "run", "teslamate-mcp", "stdio"]
    }
  }
}

Remote use (Docker)

cp env.example .env
# Set DATABASE_URL and ideally AUTH_TOKEN
docker compose up -d

The MCP endpoint is at http://localhost:8888/mcp and a liveness probe is exposed at http://localhost:8888/health.

A prebuilt multi-arch image (linux/amd64, linux/arm64) is also published to GHCR on every tagged release:

docker run --rm -e DATABASE_URL=... -p 8888:8888 ghcr.io/cobanov/teslamate-mcp:latest

Configuration

All settings are read from environment variables (.env supported). Only DATABASE_URL is required.

Variable	Default	Notes
DATABASE_URL	required	postgresql://user:pass@host:5432/teslamate
AUTH_TOKEN	empty	Enables bearer auth on the HTTP endpoint
HOST	0.0.0.0	HTTP bind host
PORT	8888	HTTP bind port
POOL_MIN_SIZE	1	psycopg pool floor
POOL_MAX_SIZE	10	psycopg pool ceiling
QUERY_TIMEOUT_MS	5000	statement_timeout for run_sql
CUSTOM_SQL_ROW_LIMIT	1000	LIMIT injected when run_sql doesn't supply one
LOG_LEVEL	INFO	Standard Python log level
DEBUG	false	Starlette debug mode (keep off in production)

Generate a bearer token:

uv run teslamate-mcp gen-token

Available tools

Predefined (18)

Vehicle: get_basic_car_information, get_current_car_status, get_software_update_history

Battery & health: get_battery_health_summary, get_battery_degradation_over_time, get_daily_battery_usage_patterns, get_tire_pressure_weekly_trends

Driving: get_monthly_driving_summary, get_daily_driving_patterns, get_longest_drives_by_distance, get_total_distance_and_efficiency, get_drive_summary_per_day

Efficiency: get_efficiency_by_month_and_temperature, get_average_efficiency_by_temperature, get_unusual_power_consumption

Charging & location: get_charging_by_location, get_all_charging_sessions_summary, get_most_visited_locations

Custom (2)

• get_database_schema — current TeslaMate schema (one row per column)
• run_sql(query) — execute a custom SELECT or WITH … SELECT

Adding a new query

1. Drop a SELECT into src/teslamate_mcp/queries/your_query.sql.

2. Add a sibling your_query.toml:

   name = "get_your_data"
   description = "What this returns."

3. Restart the server. The registry picks it up automatically.

Development

uv sync                          # install with dev deps
uv run ruff check src tests      # lint
uv run ruff format src tests     # format
uv run pytest                    # tests (Docker-backed integration tests skip if Docker is absent)

License

MIT — see LICENSE.

Featured in