Containerize MCP Server

SkillCloud & infra

Containerize an R-based MCP (Model Context Protocol) server using Docker. Covers mcptools integration, port exposure, stdio vs HTTP transport, and connecting Claude Code to the containerized server. Use when deploying an R MCP server without requiring a local R installation, creating a reproducible MCP server environment, running MCP servers alongside other containerized services, or distributing an MCP server to other developers.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Containerize MCP Server skill

What this skill tells your AI

The instructions your AI receives, as published by pjt222/agent-almanac in skills/containerize-mcp-server/SKILL.md and read by ahel’s review.

Package an R MCP server into a Docker container for portable deployment.

When to Use

  • Deploying an R MCP server without requiring a local R installation
  • Creating a reproducible MCP server environment
  • Running MCP servers alongside other containerized services
  • Distributing an MCP server to other developers

Inputs

  • Required: R MCP server implementation (mcptools-based or custom)
  • Required: Docker installed and running
  • Optional: Additional R packages the server needs
  • Optional: Transport mode (stdio or HTTP)

Procedure

Step 1: Create Dockerfile for MCP Server

FROM rocker/r-ver:4.5.0

# Install system dependencies
RUN apt-get update && apt-get install -y \
    libcurl4-openssl-dev \
    libssl-dev \
    libxml2-dev \
    libgit2-dev \
    libssh2-1-dev \
    git \
    curl \
    && rm -rf /var/lib/apt/lists/*

# Install R packages
RUN R -e "install.packages(c( \
    'remotes', \
    'ellmer' \
    ), repos='https://cloud.r-project.org/')"

# Install mcptools
RUN R -e "remotes::install_github('posit-dev/mcptools')"

# Set working directory
WORKDIR /workspace

# Expose MCP server ports
EXPOSE 3000 3001 3002

# Environment variables
ENV R_LIBS_USER=/workspace/renv/library
ENV RENV_PATHS_CACHE=/workspace/renv/cache

# Default: start MCP server
CMD ["R", "-e", "mcptools::mcp_server()"]

Expected: A Dockerfile exists in the project root with rocker/r-ver base image, system dependencies, mcptools installation, and the MCP server as the default command.

On failure: Verify the base image tag matches your R version. If remotes::install_github fails, check that git and libgit2-dev are in the system dependencies layer.

Step 2: Create docker-compose.yml

version: '3.8'

services:
  mcp-server:
    build:
      context: .
      dockerfile: Dockerfile
    container_name: r-mcp-server
    image: r-mcp-server:latest

    volumes:
      - /path/to/projects:/workspace
      - renv-cache:/workspace/renv/cache

    stdin_open: true
    tty: true

    network_mode: "host"

    environment:
      - TERM=xterm-256color
      - R_LIBS_USER=/workspace/renv/library

    restart: unless-stopped

volumes:
  renv-cache:
    driver: local

Using network_mode: "host" ensures the MCP server ports are accessible on localhost.

Expected: A docker-compose.yml file in the project root with the MCP server service, volume mounts for project files and renv cache, and stdin_open/tty enabled for stdio transport.

On failure: If volume paths are invalid, adjust /path/to/projects to the actual project directory. On Windows/WSL, use /mnt/c/... or /mnt/d/... paths.

Step 3: Build and Start

docker compose build
docker compose up -d

Expected: Container starts with MCP server running.

On failure: Check logs with docker compose logs mcp-server. Common issues:

  • Missing R packages: Add to Dockerfile RUN install step
  • Port already in use: Change exposed port or stop conflicting service

Step 4: Connect Claude Code to Container

For stdio transport (container must stay running with stdin):

claude mcp add r-mcp-docker stdio "docker" "exec" "-i" "r-mcp-server" "R" "-e" "mcptools::mcp_server()"

For HTTP transport (if the MCP server supports it):

{
  "mcpServers": {
    "r-mcp-docker": {
      "type": "http",
      "url": "http://localhost:3000/mcp"
    }
  }
}

Expected: Claude Code's MCP configuration includes the r-mcp-docker server entry, and claude mcp list shows the new server.

On failure: For stdio transport, ensure the container name matches (r-mcp-server) and that the container is running with docker ps. For HTTP transport, verify the port is exposed and reachable with curl http://localhost:3000/mcp.

Step 5: Verify Connection

# Check container is running
docker ps | grep mcp-server

# Test R session inside container
docker exec -it r-mcp-server R -e "sessionInfo()"

# Verify mcptools is available
docker exec -it r-mcp-server R -e "library(mcptools)"

Expected: docker ps shows the r-mcp-server container running, sessionInfo() returns the expected R version, and library(mcptools) loads without error.

On failure: If the container is not running, check docker compose logs mcp-server for startup errors. If mcptools fails to load, rebuild the image to ensure the package installed correctly.

Step 6: Add Custom MCP Tools

To add project-specific MCP tools, mount your R scripts:

volumes:
  - ./mcp-tools:/mcp-tools

And load them in the CMD:

CMD ["R", "-e", "source('/mcp-tools/custom_tools.R'); mcptools::mcp_server()"]

Expected: Custom R scripts are accessible inside the container at /mcp-tools/, and the MCP server loads them on startup alongside the default tools.

On failure: Verify the volume mount path is correct with docker exec -it r-mcp-server ls /mcp-tools/. If scripts fail to source, check for missing package dependencies in the custom tools.

Validation

  • Container builds without errors
  • MCP server starts inside the container
  • Claude Code can connect to the containerized server
  • MCP tools respond correctly to requests
  • Container restarts cleanly
  • Volume mounts allow access to project files

Common Pitfalls

  • stdin/tty requirements: MCP stdio transport requires stdin_open: true and tty: true
  • Network isolation: Default Docker networking may prevent localhost access. Use network_mode: "host" or expose specific ports.
  • Package versions: Pin mcptools to a specific commit for reproducibility
  • Large image size: mcptools + dependencies can be large. Consider multi-stage builds for production.
  • Windows Docker paths: When running Docker Desktop on Windows with WSL, path mapping differs

Related Skills

  • create-r-dockerfile - base Dockerfile patterns for R
  • setup-docker-compose - compose configuration details
  • configure-mcp-server - MCP server configuration without Docker
  • troubleshoot-mcp-connection - debugging MCP connectivity issues

Signals

GitHub stars
32
Forks
4
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
containerize-mcp-server
Source
github.com/pjt222/agent-almanac