BYD Vehicle Bridge — MCP Server

A secure, read-only MCP (Model Context Protocol) server that connects to your BYD electric vehicle via the BYD cloud API. Designed for AI agents (like OpenClaw, Claude Code, or any MCP client) to query real-time vehicle data — battery SOC, range, tire pressures, door states, GPS, and more — while keeping your credentials safe on your own infrastructure.

Built with Python, pyBYD, and the MCP SDK.

---

Architecture

┌─────────────────────── Host Machine ──────────────────────────┐
│                                                               │
│  ┌─────────────────────────────────────────────────────────┐  │
│  │  Docker Network: byd-internal (172.28.0.0/16)           │  │
│  │                                                         │  │
│  │  ┌──────────────────────┐    MCP (SSE)   ┌──────────┐  │  │
│  │  │  AI Agent            │◄──────────────►│  BYD     │  │  │
│  │  │  (OpenClaw / Claude) │                │  Bridge  │  │  │
│  │  │                      │                │          │  │  │
│  │  │  Tools:              │                │  Tools:  │  │  │
│  │  │  · get_battery()     │                │  · poll  │  │  │
│  │  │  · get_vehicle()     │                │  · cache │  │  │
│  │  │  · get_all_data()    │                │  · serve │  │  │
│  │  │  · get_health()      │                │          │  │  │
│  │  └──────────────────────┘                └─────┬────┘  │  │
│  │                                                 │      │  │
│  │                                        BYD Cloud API    │  │
│  │                                                 │      │  │
│  │                                            ┌────▼────┐ │  │
│  │                                            │  BYD    │ │  │
│  │                                            │  Cloud  │ │  │
│  │                                            └─────────┘ │  │
│  └─────────────────────────────────────────────────────────┘  │
│                                                               │
│  Credentials stored in .env — never leave this machine        │
└───────────────────────────────────────────────────────────────┘

Key Design Decisions

Decision	Rationale
MCP over REST	AI agents discover tools natively via the Model Context Protocol. No manual URL memorization, no raw HTTP parsing.
SSE transport	Server-Sent Events over HTTP — standard MCP transport, compatible with all MCP clients.
Background polling	The server polls the BYD API every 60s and caches results. Tools return instant cached data, never block on the network.
Internal Docker network	The bridge container exposes zero ports to the host. Only containers on the same Docker network can reach it.
Read-only by design	Remote commands (lock/unlock, AC control, windows) are intentionally excluded. This bridge reads, never writes.
Dual mode	minimal mode for privacy-sensitive users (no GPS, no door states). full mode for complete telemetry.
ghcr.io registry	Docker image published to GitHub Container Registry. Pull on any server — no rebuild needed.
PR workflow	All changes go through pull requests. Tests run automatically on every PR.

---

Features

• 🔋 Battery Monitoring — State of charge (SOC %), estimated range, charging status
• 🚗 Driving Data — Speed, power (kW), odometer, outside/cabin temperature
• 🔌 Charging Details — Voltage, current, charge rate, time to full (full mode)
• 🛞 Tire Pressures — All four wheels with units (full mode)
• 🚪 Door & Window States — Open/closed status, lock state (full mode)
• 📍 GPS Location — Latitude, longitude, heading (full mode)
• 🌡️ HVAC Status — AC on/off, target temperature, fan speed (full mode)
• 🔒 Read-Only — No remote commands. No write access. Ever.
• 🐳 Dockerized — Single container, minimal footprint, non-root user
• 🔐 Credentials Safe — BYD username/password never leave your VPS
• 🧪 28 Unit Tests — Full test suite runs on every PR via GitHub Actions
• ⚙️ CI/CD — Automated tests + Docker image build + push to ghcr.io

---

MCP Tools

Tool	Mode	Description
get_battery()	both	Battery SOC %, charging status, range, mileage, temps, speed, power
get_vehicle()	both	VIN, model, brand, plate, energy type
get_all_data()	both	Everything available (mode-dependent)
get_health()	both	Connection status, mode, poll interval, last poll time, errors

Tool Output Example

// get_battery() response
{
  "soc_percent": 73,
  "charging_status": "disconnected",
  "estimated_range_km": 280,
  "mileage_km": 12450,
  "outside_temp_c": 31.5,
  "cabin_temp_c": 28.0,
  "speed_kmh": 0,
  "engine_power_kw": 0.0,
  "fuel_percent": null,
  "last_updated": "2026-07-31T08:30:00+00:00"
}

---

Quick Start

Prerequisites

• A BYD electric vehicle with an active BYD account
• A server with Docker and Docker Compose installed
• An MCP client (OpenClaw, Claude Code, etc.)

1. Pull the Image

docker pull ghcr.io/pavelpervi/byd-vehicle-bridge:latest

2. Configure

Create a directory and .env file:

mkdir -p /opt/byd-bridge && cd /opt/byd-bridge

cat > .env << 'EOF'
BYD_USERNAME=your@email.com
BYD_PASSWORD=your-password
BYD_COUNTRY=IL
BYD_MODE=minimal   # or "full"
EOF

3. Deploy with Docker Compose

# docker-compose.yml
cat > docker-compose.yml << 'EOF'
services:
  byd-bridge:
    image: ghcr.io/pavelpervi/byd-vehicle-bridge:latest
    container_name: byd-bridge
    restart: unless-stopped
    env_file: .env
    networks:
      - byd-internal
    healthcheck:
      test: ["CMD", "python3", "-c", "import socket; s=socket.create_connection(('localhost', 8000), timeout=5); s.close()"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 15s

networks:
  byd-internal:
    driver: bridge
    ipam:
      config:
        - subnet: 172.28.0.0/16
EOF

docker compose up -d

4. Connect Your MCP Client

For OpenClaw (running on the same host):

# Connect the OpenClaw container to the bridge network
docker network connect byd-internal <openclaw-container-name>

# Register the MCP server
openclaw mcp add byd-bridge --url http://byd-bridge:8000 --transport sse --timeout 30

For Claude Code or any stdio MCP client:

{
  "mcpServers": {
    "byd-bridge": {
      "command": "docker",
      "args": ["exec", "-i", "byd-bridge", "python3", "-m", "byd_bridge"]
    }
  }
}

5. Verify

# Check the container is running
docker ps | grep byd-bridge

# View logs
docker logs byd-bridge

# Test the MCP tools
openclaw mcp call byd-bridge get_health
openclaw mcp call byd-bridge get_battery

Expected output from get_health:

{
  "status": "ok",
  "mode": "minimal",
  "vehicle_connected": true,
  "last_successful_poll": "2026-07-31T08:30:00+00:00",
  "poll_interval_s": 60,
  "error": null
}

---

Security

What's Protected

Concern	How It's Addressed
Credentials	BYD username/password stored in .env on your VPS only. Never sent to the AI agent.
Network Exposure	Zero ports published to the host. Only accessible on the internal Docker network.
Write Access	No remote commands implemented. The bridge reads data only.
Privilege Escalation	Container runs as non-root user, read-only filesystem, all Linux capabilities dropped.
GPS Privacy	GPS data only available in full mode. Default minimal mode excludes it.

Minimal vs Full Mode

Data Point	minimal	full
Battery SOC, range, charging	✅	✅
Speed, power, mileage	✅	✅
Outside/cabin temperature	✅	✅
Vehicle info (VIN, model)	✅	✅
Tire pressures	❌	✅
Door/window states	❌	✅
GPS location	❌	✅
Charging voltage/current	❌	✅
HVAC status	❌	✅

---

Configuration Reference

Variable	Default	Description
BYD_USERNAME	—	BYD account email or phone (required)
BYD_PASSWORD	—	BYD account password (required)
BYD_COUNTRY	IL	ISO country code for API region
BYD_MODE	minimal	minimal or full data scope
POLL_INTERVAL	60	Seconds between BYD API polls
BYD_PORT	8000	MCP server port (internal)

---

Project Structure

byd-vehicle-bridge/
├── src/
│   └── byd_bridge/              ← Python package
│       ├── __init__.py
│       ├── __main__.py          ← Entry: python -m byd_bridge
│       ├── config.py            ← Settings from env vars (lazy singleton)
│       ├── state.py             ← BridgeState + background poller
│       └── server.py            ← MCP server with 4 tools
├── tests/
│   ├── __init__.py
│   ├── conftest.py              ← Shared fixtures
│   ├── test_config.py           ← 12 tests: validation, defaults, errors
│   ├── test_state.py            ← 5 tests: init, data shapes, copy safety
│   └── test_server.py           ← 11 tests: before/after poll, registration
├── .github/
│   └── workflows/
│       ├── tests.yml            ← CI: runs tests on every PR
│       └── docker-build.yml     ← CD: builds & pushes image to ghcr.io
├── Dockerfile
├── docker-compose.yml
├── pyproject.toml               ← Modern Python project config
├── requirements.txt
├── .env.example
├── .gitignore
├── README.md
└── LICENSE

---

Running Tests

Tests are run automatically on every pull request via GitHub Actions, but you can also run them locally:

# Install dependencies
pip install -r requirements.txt
pip install -e .

# Run all 28 tests
python -m pytest tests/ -v

Test Coverage

Test File	Tests	What It Covers
test_config.py	12	Mode validation, env parsing, defaults, missing credentials
test_state.py	5	BridgeState init, data shapes, copy safety
test_server.py	11	Tool registration, before/after poll states, async MCP calls

---

CI/CD Pipeline

Event	Workflow	What Happens
Push to PR branch	tests.yml	Runs all 28 tests, lints with Ruff
PR merged to main	tests.yml + docker-build.yml	Tests pass, then Docker image is built and pushed to ghcr.io/pavelpervi/byd-vehicle-bridge
Tag pushed (v*.*.*)	docker-build.yml	Image tagged with semver (v1.0.0)

Retention: Only the latest 5 versions are kept in ghcr.io. Old sha-* images are automatically cleaned up.

---

How It Works

Background Polling

When the server starts, a daemon thread launches an async event loop that:

1. Authenticates to the BYD cloud API using pyBYD
2. Fetches the vehicle list and real-time data
3. In full mode, also fetches GPS, charging details, and HVAC status
4. Caches everything in memory
5. Sleeps for POLL_INTERVAL seconds, then repeats

MCP Tool Calls

When an AI agent calls a tool (e.g., get_battery()):

1. The MCP server receives the request
2. Looks up the cached data from the latest poll
3. Returns the data immediately — no network call to BYD
4. The agent receives structured, typed data

This means tool calls are instant (single-digit milliseconds) — the latency lives in the background poller, not in the agent's request.

Why SSE Transport?

MCP supports three transports:

Transport	Use Case
stdio	Local subprocess — server runs as a child of the MCP client
SSE	Remote server — HTTP with Server-Sent Events (what we use)
Streamable HTTP	Newer HTTP transport — simpler than SSE

SSE is the standard choice for a Docker-deployed MCP server. It requires no special client configuration beyond the URL.

---

Tech Stack

• Python 3.12 — Core language
• pyBYD — Async Python client for the BYD vehicle API
• MCP SDK — Model Context Protocol SDK (v2.0.0+, MCPServer class)
• Docker — Containerization
• GitHub Actions — CI/CD: tests + image build
• GitHub Container Registry — Docker image hosting
• python-dotenv — Environment variable management
• pytest — Test framework
• Ruff — Python linter

---

Development

Prerequisites

git clone https://github.com/pavelpervi/byd-vehicle-bridge.git
cd byd-vehicle-bridge
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
pip install -e .

Run the Server Locally

# Set BYD credentials
export BYD_USERNAME=your@email.com
export BYD_PASSWORD=your-password
export BYD_COUNTRY=IL
export BYD_MODE=minimal

# Start the MCP server
python -m byd_bridge

The server starts on port 8000 (or BYD_PORT if set).

Run Tests

# Set test env vars (required by config module)
export BYD_USERNAME=test@example.com
export BYD_PASSWORD=test-password

# Run all tests
python -m pytest tests/ -v

Build the Docker Image Locally

docker build -t byd-bridge .
docker run --rm -p 8000:8000 --env-file .env byd-bridge

---

Acknowledgements

• pyBYD — The excellent Python library that makes BYD API access possible
• hass-byd-vehicle — Home Assistant integration that inspired this project
• Model Context Protocol — The protocol standardizing AI-tool communication

---

License

MIT