UbiMCP composes the currently enabled biomedical MCP servers behind a single endpoint. Each server wraps a public biomedical database or API. This page lists the enabled public data sources; for routing, your LLM will use discover_servers, discover_tools, and search_tools automatically.
By category
Knowledge graph
| Server | Source | Description |
|---|
opentargets | Open Targets Platform | Target-disease associations, evidence, tractability, drugs, mechanisms of action |
Clinical and regulatory
| Server | Source | Description |
|---|
openfda | openFDA | Adverse event reports, drug labels, recalls, enforcement actions |
Gene expression
| Server | Source | Description |
|---|
gtex | GTEx Portal | Gene and isoform expression across human tissues; eQTL and sQTL |
Genomic annotation
| Server | Source | Description |
|---|
ncbi | NCBI Datasets | Genomes, genes, orthologs, taxonomy |
mygene | MyGene.info | Unified gene annotation across multiple sources |
Chemistry and drugs
| Server | Source | Description |
|---|
mychem | MyChem.info | Unified chemical and drug annotations |
Disease annotation
| Server | Source | Description |
|---|
mydisease | MyDisease.info | Unified disease annotations and cross-references |
Protein and antibody records
| Server | Source | Description |
|---|
antibody-registry | Antibody Registry | Searchable registry of research antibodies with RRIDs |
Discovery from the client
You rarely need to memorize server names. Inside any LLM session, the model can call:
discover_servers(scope="ubi-bio") to enumerate enabled biomedical servers with descriptions.discover_servers(category="chemical") to filter by category.discover_tools("opentargets") to see what tools that server exposes.search_tools("eQTL") to find tools across all servers matching a phrase.get_tool_schema(tool_name) to see the exact parameters before calling.
This pattern is documented in detail on Discovery Tools.
The enabled servers expose many database-specific tools in total. The composition server presents a unified execute_tool interface so that an LLM only ever sees six core tools regardless of how many servers are enabled. This is what keeps context usage low.