← all posts
ShipMarch 8, 2026·5 min read

MCP Server via Docker: Five Tools, One Pull, Zero Config Hell

The hardest part of running an MCP server has always been the Python environment. PATH issues, wrong venvs, 'server disconnected' at 2am. We packaged everything into a Docker image. One pull. One config line. Done.

Every MCP developer has hit it. You configure the server, restart Claude Desktop, and get: MCP agenticstore: Server disconnected. You check your terminal — it works fine. The client just can't find your Python, your uvx, or the environment where you installed everything.

The root cause is always the same: GUI applications like Claude Desktop launch with a stripped-down environment. They don't know about your shell's PATH, your active virtualenv, or which Python binary you're actually using. Managing this is a tax every MCP user pays — until now.

Today's Docker image eliminates the problem entirely. The image bundles Python, every dependency, and all five tools into a single artifact. Your AI client runs a container. No PATH issues. No venv confusion. No version drift. One pull and you're done.

Before Docker

✗Install Python — which version?
✗Install uv — add to PATH manually
✗Client can't find uvx (different environment)
✗Works in terminal, broken in Claude Desktop
✗'Server disconnected' at 2am
✗Update = reinstall and reconfigure

After Docker

✓docker pull agenticstore/agentic-store-mcp:latest
✓Paste one JSON config
✓Restart your AI client
✓All 31 tools ready
✓Update = docker pull
✓Works identically on macOS, Linux, Windows

Get all five tools in two commands

bash
# Pull the image
docker pull agenticstore/agentic-store-mcp:latest

# Verify — should list all 5 tools
docker run --rm agenticstore/agentic-store-mcp:latest --list

Then add this to your AI client config and restart:

claude_desktop_config.json · cursor mcp.json · windsurf mcp_config.json
{
  "mcpServers": {
    "agenticstore": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "agenticstore/agentic-store-mcp:latest"]
    }
  }
}

The same config works in Claude Desktop, Cursor, Windsurf, and VS Code. The quickstart guide now defaults to Docker, with step-by-step install instructions for every OS.

One server. Five tools. Everything bundled.

The Docker image ships every tool in a single server process — you don't manage five separate MCP configs or five separate processes. One entry in your AI client config gives you the full toolkit.

AgenticCode3 tools · zero external deps
python_lint_checker

Catch bugs and style violations before your agent ships broken Python.

repo_scanner

Detect leaked API keys, credentials, PII, and .gitignore gaps. 40+ patterns.

dependency_audit

Scan requirements.txt, package.json, go.mod, Cargo.toml against the OSV CVE database.

AgenticData2 tools · live web access
agentic_web_crawl

Fetch any URL — title, meta, OG tags, headings, links. Works out of the box, no config.

agentic_web_search

Real-time internet search via SearXNG. Optional — needs docker compose up for SearXNG.

Always on the latest version

Every time we ship new tools or fixes, staying current is one command:

bash
docker pull agenticstore/agentic-store-mcp:latest

We tag every release. :latest always points to the most recent stable build. Pin to a specific version (e.g. :0.4.0) if you need reproducibility in CI.

The foundation for more advanced tools

Docker isn't just a distribution convenience — it's what makes the next wave of tools possible. Some of what we're building requires system libraries, compiled binaries, or native extensions that are painful to distribute via PyPI. A container image removes that constraint entirely.

Expect more AgenticCode and AgenticData tools shipping on this cadence. Follow ships here on the blog or watch the GitHub repo for releases.

Ready to end setup hell?

Docker. One pull. Five tools. Works on macOS, Linux, and Windows with any MCP-compatible AI client.

Get started in 2 minutes →

Frequently asked questions

← all posts
read the docs →