Has anyone used docker with ADK with MCP servers?

[xuanyang15]

Hi 👋

I have gotten MCP servers working with ADK but I am trying to figure out to make it work using docker so that the ADK app can be deployed anywhere.

From what i understand now, best practice would be to use docker compose so that each MCP server is a separate container.
I also found out that docker desktop has their own MCP Toolkit now which seems to be great.

Search online, this depo, and the subreddit, docker does not really seem to be something that is used right now when it comes to ADK. Or it could just be that people that are learning it (like me) does not have the skills yet to implement it.

[xuanyang15]

Hi! I have have docker containers running MCP servers locally.

This is my docker compose file

# Set working directory
WORKDIR /app

# Install system dependencies including Node.js (latest)
RUN apt-get update && apt-get install -y \
    git \
    curl \
    && curl -fsSL https://deb.nodesource.com/setup_current.x | bash - \
    && apt-get install -y nodejs \
    && rm -rf /var/lib/apt/lists/*

# Copy requirements first for better caching
COPY requirements.txt .

# Install uv for faster dependency resolution
RUN pip install --upgrade pip uv

# Install Python dependencies using uv (much faster than pip)
RUN uv pip install --system --no-cache -r requirements.txt

# Copy the entire backend directory contents to /app
COPY . ./

# Copy any .env files from parent directory if they exist (using shell form)
RUN cp ../.env* ./ 2>/dev/null || true

# Set Python path
ENV PYTHONPATH=/app

# Set environment variables
ENV GOOGLE_GENAI_USE_VERTEXAI=False

# Default command (will be overridden)
CMD ["python", "scripts/process_single_job.py"] 

[xuanyang15]

Thanks mate! This will be very helpful.

Can you explain what this line does:

Copy any .env files from parent directory if they exist (using shell form)

RUN cp ../.env* ./ 2>/dev/null || true

Does it mean that if you have a .env file in your project root then it gets copied?

Do you know how much this changes if you use gcloud instead?

Also, i read that you can (perhaps should) install each mcp server as its own seperate service. Kind of like this (dont know if this works, just an example file i have not tested):

version: '3.8'
services:
fastapi-app:
build: .
ports:
- "8080:8080"
depends_on:
- mcp-server

mcp-server:
image: mongodb/mongodb-mcp-server:latest
environment:
- MDB_MCP_CONNECTION_STRING=mongodb://mongo:27017
ports:
- "8081:8080" # MCP server runs on 8080 by default, mapped to 8081 on host
depends_on:
- mongo

mongo:
image: mongo:6.0
restart: unless-stopped
ports:
- "27017:27017"
volumes:
- mongo_data:/data/db

volumes:
mongo_data:

[xuanyang15]

Hi 👋

I have gotten MCP servers working with ADK but I am trying to figure out to make it work using docker so that the ADK app can be deployed anywhere.

From what i understand now, best practice would be to use docker compose so that each MCP server is a separate container. I also found out that docker desktop has their own MCP Toolkit now which seems to be great.

Search online, this depo, and the subreddit, docker does not really seem to be something that is used right now when it comes to ADK. Or it could just be that people that are learning it (like me) does not have the skills yet to implement it.

So I have done this.

But is doesn't work on windows, becuase spawning process in windows throws notimplemented exception.

You have to create a python script to start univicorn server, and set the

if sys.platform == "win32":
asyncio.set_event_loop_policy(asyncio.WindowsProactorEventLoopPolicy())

and use this script instead of adk web

I hope this issue is addressed more correctly.

[xuanyang15]

Response from ADK Answering Agent

TLDR: Running MCP servers as separate services in Docker containers is a good approach. For deployment to Google Cloud, you should manage environment variables through the cloud service's configuration rather than copying .env files. The ADK MCPToolset is used to connect your agent to these separate MCP services.

---

Hi! I can provide some information on your questions about using Docker with ADK and MCP servers.

Explanation of the Docker command

You asked about the line RUN cp ../.env* ./ 2>/dev/null || true.

This command is used to copy any files starting with .env from the parent directory of your Docker build context into the /app directory inside the container. This is a common pattern for local development to easily include environment variables for things like API keys. [20] The 2>/dev/null || true part ensures that the Docker build doesn't fail if no .env files are found.

Changes when using gcloud for deployment

When you deploy your application to Google Cloud using gcloud, the way you handle environment variables changes. Instead of copying a .env file into your container, it is best practice to manage secrets and environment variables using the deployment environment's features.

• Deployment Options: You can deploy your ADK agent to various Google Cloud services, including Cloud Run and Google Kubernetes Engine (GKE). [3]
• Cloud Run: You can deploy using the adk deploy cloud_run command or gcloud run deploy. With either method, you can set environment variables directly in the service configuration. [2]
• GKE: You can deploy using adk deploy gke or by manually applying Kubernetes manifests with kubectl. The manifests will define the environment variables for your container. [1]

In both cases, you would remove the COPY ../.env* ./ line from your Dockerfile for production builds and instead inject the configuration at deployment time.

MCP server as a separate service

Your understanding is correct! Running MCP servers as separate services is a recommended pattern. The Model Context Protocol (MCP) is designed around a client-server architecture. [5]

Your ADK agent acts as an MCP client and can connect to one or more MCP servers. The MCPToolset class in ADK is specifically designed for this purpose. It can connect to a server running in a separate process, which could be a Docker container managed by Docker Compose as in your example, or a remote server. [4]

Your docker-compose.yml example is a great illustration of how to set up a multi-container application with an ADK agent and a separate MCP server. This is a solid architecture for both local development and production deployments.

[4] https://github.com/google/adk-docs/blob/main/docs/tools/mcp-tools.md
[5] https://github.com/google/adk-docs/blob/main/docs/mcp/index.md
[20] https://github.com/google/adk-docs/blob/main/examples/python/tutorial/agent_team/adk-tutorial/readme.md
[3] https://github.com/google/adk-docs/blob/main/docs/deploy/index.md
[2] https://github.com/google/adk-docs/blob/main/docs/deploy/cloud-run.md
[1] https://github.com/google/adk-docs/blob/main/docs/deploy/gke.md