Docker image for installing and running tools for LLM agents (MCP, OpenAPI, UVX, NPX, Python)

• Python / Node.js / Deno runtime, Rust toolchain - includes python, node, uvx, npx, deno, cargo
• Includes extra packages for managing MCP/OpenAPI tools and connections
  • mcpo - MCP to OpenAPI bridge
  • supergateway - MCP STDIO/SSE bridge
  • @modelcontextprotocol/inspector - debugging tool for MCP
• Utils: curl, jq, git, ffmpeg
• Easy unified cache at /app/cache for all tools
• Scanned with trivy for vulnerabilities

The agentic variant includes Python packages for common agent tasks:

• File formats — PyPDF2, python-docx, openpyxl, Pillow, lxml, beautifulsoup4, markdownify, chardet
• Data analysis — pandas, numpy, matplotlib, tabulate
• Data access — requests, httpx, aiohttp, sqlalchemy, psycopg2-binary
• Essentials — pydantic, tiktoken

# Launch MCP tools in stdio mode
docker run ghcr.io/av/tools uvx mcp-server-time

# Bridge from MCP to OpenAPI
docker run -p 8000:8000 ghcr.io/av/tools uvx mcpo -- uvx mcp-server-time --local-timezone=America/New_York
# http://0.0.0.0:8000/docs -> see endpoint documentation

# Run MCP inspector
docker run -p 6274:6274 -p 6277:6277 ghcr.io/av/tools npx @modelcontextprotocol/inspector

# Persist the cache volume for quick restarts
# -v cache:/app/cache - named docker volume
# -v /path/to/my/cache:/app/cache - cache on the host
docker run -v cache:/app/cache ghcr.io/av/tools uvx mcp-server-time

# Use the agentic variant for Python-heavy workflows
docker run ghcr.io/av/tools:agentic python3 -c "import pandas; print(pandas.__version__)"

In docker compose:

services:
  time:
    image: ghcr.io/av/tools
    command: uvx mcp-server-time
    volumes:
      - cache:/app/cache

  fetch:
    image: ghcr.io/av/tools
    command: uvx mcpo -- uvx mcp-server-fetch
    ports:
      - 7133:8000
    volumes:
      - cache:/app/cache

Check out Harbor for a complete dockerized LLM environment.