INSTALLATION.md

📦 Installation Guide

Option A: Docker Compose (Recommended)

A1. Development mode (recommended for now)

Builds from the local Dockerfile and mounts ./dolphin_mcp_pilot into the container as read-only. Source changes require a container restart (uvicorn is not started with --reload):

# 1. Prepare environment
cp .env.example .env
# edit .env — at minimum set DS_URL and DS_TOKEN (see Configuration below)

# 2. Start (dev profile builds locally)
docker compose --profile dev up -d dolphin-mcp-pilot-dev

# 3. Verify
docker compose --profile dev ps        # STATUS should show (healthy) after ~10s
docker compose --profile dev logs -f   # watch startup logs

The service will be available at http://localhost:8001/mcp/ (note the trailing slash).

A2. Production mode (published image)

Note: The ghcr.io/iflytek/dolphin-mcp-pilot:latest image is published automatically when a stable release tag (e.g. v0.2.0) is pushed. Until the first release is tagged, use A1 or Option B/C below.

Once a release is available, the prod service pulls from ghcr.io:

# Optional: pin to a specific version
echo "IMAGE_TAG=0.2.0" >> .env

docker compose up -d
docker compose ps        # STATUS should show (healthy)

See docker-compose.yml for the full reference (resource limits, log rotation, healthcheck, etc.).

Option B: From source

git clone https://github.com/iflytek/dolphin-mcp-pilot.git
cd dolphin-mcp-pilot
pip install -r requirements.txt
python -m dolphin_mcp_pilot

Option C: Install as package

pip install .
dolphin-mcp-pilot

🏃 Run Modes

stdio mode (for local AI tools)

python -m dolphin_mcp_pilot

HTTP mode (for remote AI agents)

DS_MCP_TRANSPORT=http python -m dolphin_mcp_pilot

Or with Docker:

docker compose up -d

📦 Packaging & Distribution

Build wheel:

pip install build
python -m build

Install locally:

pip install .

Then run:

dolphin-mcp-pilot

---

← Back to README | Features | Configuration | Deployment