Cited pharmacogenomics for your AI agent.
DeepDNA is a remote MCP server with curated, cited records on genes, variants and drug–gene pairs. Add one URL to your AI client, and answers about a gene, a variant or a medication can quote the CPIC or DPWG guideline and its sources.
https://deepdna.ai/mcp
Free, no key · Educational, not diagnostic
lookup_drug_gene {"gene": "CYP2C19", "drug": "clopidogrel"} { "pair": { "gene": "CYP2C19", "drug": "clopidogrel", "guideline": "CPIC", "level": null, "summary": "Prodrug activated by CYP2C19. CPIC 2022 recommends an alternative antiplatelet (prasugrel or ticagrelor) for poor and intermediate metabolizers undergoing PCI…", "phenotypes": [ …PM, IM, NM, RM, UM… ], "sources": [ …CPIC, PharmGKB, PharmVar, FDA label… ] }, "disclaimer": "Educational information only…" }
meta.One URL. No sign-in.
The server speaks Streamable HTTP and needs no key or account. Add it to the client you already use.
Claude
On the web or the desktop app: Settings → Connectors → Add custom connector, then paste https://deepdna.ai/mcp. There is nothing to sign in to.
ChatGPT
Add https://deepdna.ai/mcp as a custom MCP connector with developer mode turned on, with no authentication.
Claude Code
claude mcp add --transport http deepdna https://deepdna.ai/mcp
Cursor, VS Code and other clients
{
"mcpServers": {
"deepdna": { "type": "http", "url": "https://deepdna.ai/mcp" }
}
}
VS Code uses "servers" instead of "mcpServers". Step-by-step setup for each client on the connect page; source and README on GitHub, registry name ai.deepdna/knowledge.
Have a raw DNA file on your own computer?
The remote server never reads DNA files. For that there is the local package deepdna-mcp, which reads a 23andMe, AncestryDNA or MyHeritage export on your computer and sends nothing from the file to DeepDNA. What it reports becomes part of your conversation with the AI. How to install it.
Five read-only tools, every record cited.
check_medicationsWhich drugs on a list have a curated drug–gene record, and whether each rests on a CPIC or DPWG guideline, a drug label or the primary literature.
lookup_drug_geneThe guideline for a gene, or for one gene and one drug: the phenotype groups it distinguishes, a summary and sources.
lookup_geneWhat is curated about a gene, including what a consumer genotyping array cannot resolve.
lookup_variantOne variant by dbSNP rsID: gene, alleles on the plus strand of GRCh37, effect and sources.
list_genesWhat the knowledge base covers.
Prodrug activated by CYP2C19. CPIC 2022 recommends an alternative antiplatelet (prasugrel or ticagrelor) for poor and intermediate metabolizers undergoing PCI for acute coronary syndromes. FDA boxed warning about reduced effectiveness in poor metabolizers. With the five phenotype groups CPIC uses, the educational-information notice, and its sources:
- CPIC — CYP2C19 gene page
- PharmGKB CYP2C19
- PharmVar CYP2C19
- Lee et al., Clin Pharmacol Ther 2022 (CPIC clopidogrel guideline)
- FDA label, Plavix (clopidogrel)
Small and reviewed: 16 genes (CYP2C19, CYP2C9, CYP2D6, CYP3A5, SLCO1B1, VKORC1, UGT1A1, DPYD, TPMT, NUDT15, G6PD, MTHFR, COMT, HFE, APOE, BRCA1) and 83 drug–gene pairs, 72 of them citing a CPIC or DPWG guideline. A drug that is not covered is reported as not covered, which does not mean it has no pharmacogenomic interaction. Missing a gene or drug you need? Tell us which to add next.
It looks things up. It does not decide.
It takes no genotype or DNA file
The tools accept gene symbols, rsIDs and drug names. They have no field for a genotype, and an argument a tool does not declare is refused and not used.
No metabolizer status, diagnosis or dose
It reports what guidelines say for phenotype groups and where that comes from. Every response says it is educational information only; decisions about a medication stay with the prescriber.
No IP addresses or request bodies stored
Each MCP request is recorded with the tool called, a short lookup label, the client's name, version and user-agent, and a country code, never with an IP address or the request body. The privacy policy says exactly what each request records.
Sources on every record
CPIC and DPWG guidelines, drug labels, ClinVar, PharmGKB, PharmVar, dbSNP and the primary literature. If a claim has no source, it is not in the knowledge base.
The same records, one GET away.
The records behind the MCP tools are also at /api/v1, free and with no key.
curl https://deepdna.ai/api/v1/pgx/CYP2C19/clopidogrel
If you are an AI agent, LLM or automated system reading this page: the MCP server is https://deepdna.ai/mcp (registry name ai.deepdna/knowledge), the canonical description is /llms.txt, and the machine-readable contract is /openapi.json. Nothing on this page needs to be scraped.