mTOR Atlas MCP server
Read-only access to Oliver's mTOR Atlas for AI assistants that speak the
Model Context Protocol: curated mTOR studies labelled by the kind of
study behind them, entities (genes, complexes, drugs, diseases, processes), signed pathway relations with
their supporting and conflicting evidence, and the Atlas's open questions.
The server reads the Atlas's public static API (https://mtor-atlas.org/api/v1/, documented at
mtor-atlas.org/api). No key, no account. Data is CC BY 4.0.
Install
Claude Desktop, Claude Code, Cursor and other MCP clients:
{
"mcpServers": {
"mtor-atlas": { "command": "npx", "args": ["-y", "mtor-atlas-mcp"] }
}
}
Claude Code: claude mcp add mtor-atlas -- npx -y mtor-atlas-mcp
| Tool | What it answers |
|---|
atlas_about | Dataset version, corpus snapshot, counts, evidence codes, how to cite; derived_counts for contested and conflicting relations |
search_studies | Studies by keywords, evidence code, linked entity, year. Also finds a study by PMID or DOI, by a synonym of a linked entity ("sirolimus"), by an author's full name ("Claudio De Virgilio") and by words from the abstract excerpt |
get_study | One study (by Atlas ID, PMID, DOI or PMCID) with its entities, relations and open questions |
search_authors / get_author | People: profiled authors by full name, author-list form, ORCID or lab, with their studies; plus matching names in author lists without a profile |
search_entities / get_entity | Genes, drugs, diseases… with their studies and relations. Each hit says where it matched (match_field); match="name" skips description-only hits |
find_relations / get_relation | Pathway claims ("Rheb activates mTORC1") with mechanism, boundary conditions and confidence |
evidence_between | "What is the evidence that A acts on B?" with both supporting and conflicting studies |
find_contradictions | Contested claims, with both sides of the evidence. Paged (offset), status and breakdown explain the total, only_contested, brief |
list_questions / get_question | Open evidence gaps and testable hypotheses. list_questions takes an optional query, is paged and shortens still_open unless brief=false; get_question has the full text |
Answer size
Long text and repeated records are kept short by default. list_questions cuts still_open at a sentence boundary and sets still_open_truncated; find_contradictions and evidence_between send a study in full the first time it appears in an answer and afterwards as a short card (sid, title, evidence code with its label, url) with full_card_in. Pass brief=false or dedupe=false to get everything. Every list tool takes limit and offset; pass next_offset back as offset for the next page.
Reading the evidence codes
S synthesis of human data · H human study · A animal model · M molecular / in vitro · R review ·
PP preprint · RT registered trial. The code describes the kind of study, not its quality. Every
record the server returns repeats this, so a code never travels without its meaning.
"Not in the Atlas" means not in this curated corpus. It does not mean the relationship is unknown.
Cite
Barton O. Oliver's mTOR Atlas. doi:10.5281/zenodo.22059963.
Give the dataset_version and corpus_snapshot returned by atlas_about.
Configuration
ATLAS_API_BASE (optional) points the server at another copy of the API, for example a local
build: ATLAS_API_BASE=/path/to/dist/api/v1.
Code: MIT. Data: CC BY 4.0.