CERN Open Data MCP. Query 66k LHC Datasets by Topic, DOI, or Experiment

CERN Open Data MCP Server connects your AI client straight to the world's biggest particle physics dataset repository. Forget searching through messy web portals; you get direct access to 66,000+ datasets from Large Hadron Collider and LEP experiments. This isn't just a search engine—it's a full data workflow toolset.

First things first: You gotta make sure the connection is good before running anything big. Use check_cern_opendata_status to verify that your AI client can talk to the CERN Open Data API, so you don't waste time on dead ends.

To get oriented in this massive data jungle, start broad. You can run get_portal_statistics to see high-level stats across the whole repository, checking how many records exist by type and year. If you need context for your report, get_glossary pulls up definitions for technical jargon—you know, terms like 'luminosity' or 'b-tagging.' You can also get a map of the available research areas using list_categories, which lists every physics category and subcategory while telling you exactly how many datasets are in each bucket.

To see which major collaborations contributed data, run list_experiments to list every experiment like CMS, ATLAS, or ALICE, along with their dataset counts.

When you're ready to drill down into the actual data, you have a few ways to filter your search results. You can focus on specific physics topics using search_by_category, filtering by things like 'Exotica' or 'Higgs Physics.' If your research hinges on the collision parameters, narrow it down first with search_by_collision_type (like proton-proton 'pp' or electron-positron 'e+e-'), then use search_by_collision_energy to specify a precise energy like 13 TeV or 7 GeV.

You can also limit your search by targeting one specific collaboration using search_by_experiment. For the deepest dive, run search_datasets; this performs a full-text query across all datasets and lets you combine filters for experiment, energy, and collision type simultaneously.

Finding specific records is super easy. If you only have a publication's Digital Object Identifier (DOI), use get_record_by_doi to find the associated open dataset record immediately. Alternatively, if you have a record ID, get_record pulls all the detailed metadata for that dataset—you get abstracts, authors, file counts, and direct links right off the bat.

Once you've got the metadata, you can check exactly what files are in play using list_record_files; this returns a list of every physical file associated with the record, giving you the size, checksum, and the specific URI you need for download.

When you’re writing up your project, you won't just be dealing with data records. You can use search_documentation to find technical guides and policies about the dataset structure or analysis workflow itself. Need code? Run search_software to locate reconstruction software and specific analysis frameworks used alongside this data. If you gotta reproduce a published study, search_supplementaries finds those supplementary materials and configuration files required for accurate reproduction.

And don't forget context. You can get an overall picture of the repository using get_portal_statistics, which gives high-level stats broken down by experiment type and data year. If you need general information about a specific dataset record, even if you don't have a DOI, running get_record with a simple ID provides all the necessary metadata to get started.