Folklore MCP

Folklore MCP is a remote server you add to an AI client as a URL over Streamable HTTP, so there's nothing to install locally. Once it's connected, the client can call 11 read-only tools with no account or API key. You can ask which genes have curated evidence for Marfan syndrome, which diseases are associated with FBN1, or how a specific variant classifies under the ACMG/AMP guidelines. The classification comes back with its criterion evidence, source citations and limitations.

Variant queries accept GRCh38 coordinates, selected HGVS forms, SPDI and rsIDs, covering nuclear germline SNVs and simple indels under 50 base pairs. Gene lookups start by resolving a symbol such as BRCA1 to a stable identifier and snapshot, and later calls reuse that snapshot so the answers stay tied to the same data version. The setup page has configuration snippets for ChatGPT, Claude, Codex, Cursor and VS Code.

The anonymous tools fit researchers, bioinformaticians and developers who already work with variant nomenclature and want an AI assistant to fetch sourced evidence instead of answering from memory. The documentation keeps a lot of technical detail in view, from cursors to snapshot IDs, and it assumes you know what an ACMG criterion is. Patients trying to read their own results aren't the audience, and the project asks that qualified professionals review output before any clinical use.

With a Folklore API key, the same endpoint exposes 29 tools, including submitting case analyses from files in S3, checking their status and querying the frozen result exports. Those runs count against subscription capabilities, usage limits and credits.

The anonymous tools take public identifiers, variant expressions and literature searches, and the project says to use public identifiers only. For case work, Folklore's data and privacy documentation says private genomic workloads run on Helena-controlled infrastructure in the EEA, that AI-assisted features can send a limited set of case-derived findings or phenotype terms to external model providers, and that processing, retention and deletion for laboratories are set out in the customer agreement and DPA. The same page points out that removing names doesn't make genomic data anonymous.