atlasopenmagic-mcp

MCP Server for ATLAS Open Data metadata and file retrieval.

This server wraps the atlasopenmagic Python library as an MCP (Model Context Protocol) server, allowing LLMs to discover, query, and retrieve ATLAS Open Data datasets.

Architecture

LLM <--MCP/stdio--> atlasopenmagic-mcp serve <--HTTPS--> ATLAS Open Data API
                                                          (atlasopenmagic-api.app.cern.ch)

No authentication is required — all ATLAS Open Data is publicly accessible.

Related MCP server: CERN GitLab MCP Server

Installation

pip install atlasopenmagic-mcp

Or with pixi:

pixi install

Usage

As an MCP server

atlasopenmagic-mcp serve

Claude Desktop configuration

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "atlas-open-data": {
      "command": "atlasopenmagic-mcp",
      "args": ["serve"]
    }
  }
}

Available Tools

Available Resources

Security and data

• Read-only. No tool writes, deletes or submits data. atlas_set_release only changes the active release in the session.

• No authentication. The server uses public ATLAS Open Data only.

• No personal data. The server stores nothing and has no logging code for user input.

• Network: the server calls atlasopenmagic-api.app.cern.ch over HTTPS through the atlasopenmagic library. It cannot fetch arbitrary URLs. It returns file URLs but does not download them.

• Tool output is data, not instructions. An agent must not follow text found in dataset metadata.

• Each response is capped at 2 MiB.

• The streamable-http transport has no authentication. Do not expose it to untrusted networks.

See SECURITY.md, TOOL_CARD.md and CHANGELOG.md.

Development

# Run tests
pixi run test

# Run linting
pixi run lint

# Run all checks
pixi run check-all

Keeping in sync with atlasopenmagic

This server depends on atlasopenmagic as a pip dependency. When the upstream library is updated:

1. Non-breaking changes (new fields, new releases): work automatically — the MCP tools call through to atlasopenmagic functions, so new data surfaces without any changes here.

2. New functions: add a new tool module or extend an existing one, register it in server.py.

3. Breaking changes (renamed/removed functions, changed signatures): update the affected tool modules to match the new API, then bump the version pin in pyproject.toml.

To bump the dependency:

# In pyproject.toml, update:
#   "atlasopenmagic>=1.9"  →  "atlasopenmagic>=1.10"

License

Apache-2.0