@omophub/omophub-mcp

OMOPHub MCP Server

Medical vocabularies for AI agents.
Search, map, and navigate 10M+ OMOP concepts: SNOMED CT, ICD-10, RxNorm, LOINC, and more. Directly from Claude, Cursor, VS Code, or any MCP-compatible client.

Quick Start · Tools · Examples · Docs · Website

Why OMOPHub MCP?

Working with medical vocabularies today means downloading multi-gigabyte CSV files, loading them into a local database, and writing SQL to find what you need. Every time.

OMOPHub MCP Server gives your AI assistant instant access to the entire OHDSI ATHENA vocabulary. No database setup, no CSV wrangling, no context switching. Just ask.

You: "Map ICD-10 code E11.9 to SNOMED"

Claude: Found it - E11.9 (Type 2 diabetes mellitus without complications)
        maps to SNOMED concept 201826 (Type 2 diabetes mellitus)
        via standard 'Maps to' relationship.

Use cases:

• Concept lookup - Find OMOP concept IDs for clinical terms in seconds
• Cross-vocabulary mapping - Map between ICD-10, SNOMED, RxNorm, LOINC, and 100+ vocabularies
• Hierarchy navigation - Explore ancestors and descendants for phenotype definitions
• Concept set building - Let your AI agent assemble complete concept sets for cohort definitions
• Code validation - Verify medical codes and check their standard mappings

Quick Start

1. Get an API Key

Sign up at omophub.com → create an API key in your dashboard.

2. Add to Your AI Client

Claude Desktop

Open Claude Desktop settings > "Developer" tab > "Edit Config". Add to claude_desktop_config.json:

{
  "mcpServers": {
    "omophub": {
      "command": "npx",
      "args": ["-y", "@omophub/omophub-mcp"],
      "env": {
        "OMOPHUB_API_KEY": "oh_your_key_here"
      }
    }
  }
}

Claude Code

claude mcp add omophub -- npx -y @omophub/omophub-mcp
# Then set OMOPHUB_API_KEY in your environment

Cursor

Open the command palette and choose "Cursor Settings" > "MCP" > "Add new global MCP server". Add to .cursor/mcp.json:

{
  "mcpServers": {
    "omophub": {
      "command": "npx",
      "args": ["-y", "@omophub/omophub-mcp"],
      "env": {
        "OMOPHUB_API_KEY": "oh_your_key_here"
      }
    }
  }
}

VS Code

Add to .vscode/mcp.json:

{
  "servers": {
    "omophub": {
      "command": "npx",
      "args": ["-y", "@omophub/omophub-mcp"],
      "env": {
        "OMOPHUB_API_KEY": "oh_your_key_here"
      }
    }
  }
}

Streamable HTTP (Remote / Hosted)

Run the MCP server as an HTTP service that clients connect to via URL:

# Start HTTP server on port 3100
npx -y @omophub/omophub-mcp --transport=http --port=3100 --api-key=oh_your_key_here

# MCP endpoint: http://localhost:3100/mcp
# Health check:  http://localhost:3100/health

Connect MCP clients to / or /mcp. Useful for centralized deployments where multiple AI agents share one server instance.

Hosted (mcp.omophub.com)

Connect directly to the OMOPHub-hosted MCP server - no installation required. Each client authenticates with their own API key via the Authorization header:

Claude Code:

claude mcp add omophub --transport http \
  -H "Authorization: Bearer oh_your_key_here" \
  https://mcp.omophub.com

VS Code (.vscode/mcp.json):

{
  "servers": {
    "omophub": {
      "type": "http",
      "url": "https://mcp.omophub.com",
      "headers": { "Authorization": "Bearer oh_your_key_here" }
    }
  }
}

Cursor / Windsurf:

{
  "mcpServers": {
    "omophub": {
      "url": "https://mcp.omophub.com",
      "headers": { "Authorization": "Bearer oh_your_key_here" }
    }
  }
}

Note: Claude Desktop's Custom Connectors UI only supports OAuth and cannot send custom headers. Use the npx setup instead.

Docker

# HTTP mode (default in Docker) - serves MCP on port 3100
docker run -e OMOPHUB_API_KEY=oh_your_key_here -p 3100:3100 omophub/omophub-mcp

# Stdio mode (for piping)
docker run -i -e OMOPHUB_API_KEY=oh_your_key_here omophub/omophub-mcp --transport=stdio

3. Start Asking

"What's the OMOP concept ID for type 2 diabetes?"

"Map ICD-10 code E11.9 to SNOMED"

"Show me all descendants of Diabetes mellitus in SNOMED"

Available Tools

Tool	What it does
search_concepts	Search for medical concepts by name or clinical term across all vocabularies
get_concept	Get detailed info about a specific OMOP concept by concept_id
get_concept_by_code	Look up a concept using a vocabulary-specific code (e.g., ICD-10 E11.9)
map_concept	Map a concept to equivalent concepts in other vocabularies (paginated — see note below)
get_hierarchy	Navigate concept hierarchy - ancestors, descendants, or both
list_vocabularies	List available medical vocabularies with statistics
semantic_search	Search using natural language with neural embeddings (understands clinical meaning)
find_similar_concepts	Find concepts similar to a reference concept, name, or description
explore_concept	Get concept details, hierarchy, and cross-vocabulary mappings in one call
fhir_resolve	Resolve a FHIR coded value (incl. administrative codes via the HL7 FHIR-to-OMOP IG ConceptMaps) to its OMOP standard concept and CDM target table
fhir_resolve_codeable_concept	Resolve a FHIR CodeableConcept — best match by OHDSI vocabulary preference, honoring userSelected

Resources

URI	Description
omophub://vocabularies	Full vocabulary catalog with statistics
omophub://vocabularies/{vocabulary_id}	Details for a specific vocabulary

Prompts

Prompt	Description
phenotype-concept-set	Guided workflow to build a concept set for a clinical phenotype
code-lookup	Look up and validate a medical code with mappings and hierarchy

Example Prompts

Find a concept → search_concepts

"Search for metformin in RxNorm"

Cross-vocabulary mapping → map_concept

"I have SNOMED concept 201826 - what's the ICD-10 code?"

Build a concept set → search_concepts → get_hierarchy → map_concept

"Help me build a concept set for Type 2 diabetes including all descendants"

Completeness matters here. map_concept returns one page at a time (page_size 1-200, default 100). A widely-used drug ingredient can have well over a thousand NDC mappings, so a single call is a sample, not the answer. Check has_more in the response and keep incrementing page until it is false — otherwise the concept set looks complete while silently missing codes.

Use find_similar_concepts for exploration, not for this: it is ranked embedding similarity, so it has no notion of set membership and cannot be exhaustive no matter how large page_size is.

Validate a code → get_concept_by_code → map_concept

"Is ICD-10 code E11.9 valid? What does it map to in SNOMED?"

Semantic search → semantic_search

"Find concepts related to 'heart attack'"

Explore a concept → explore_concept

"Give me everything about SNOMED concept 201826"

FHIR-to-OMOP resolution → fhir_resolve

"Resolve FHIR SNOMED code 44054006 to OMOP — what table does it go in?"

CodeableConcept → fhir_resolve_codeable_concept

"This CodeableConcept has both SNOMED 44054006 and ICD-10 E11.9 — which should I use for OMOP?"

Find similar → find_similar_concepts

"What concepts are similar to 'Type 2 diabetes mellitus'?"

Configuration

Environment Variables

Variable	Required	Description
OMOPHUB_API_KEY	✅	Your OMOPHub API key
OMOPHUB_BASE_URL		Custom API base URL (default: https://api.omophub.com/v1)
OMOPHUB_LOG_LEVEL		debug · info · warn · error (default: info)
OMOPHUB_ANALYTICS_OPTOUT		Set to true to disable analytics headers
MCP_TRANSPORT		stdio (default) or http
MCP_PORT		HTTP server port (default: 3100, only used with http transport)
HEALTH_PORT		Port for standalone health endpoint in stdio mode (default: disabled)

CLI Arguments

# Stdio mode (default)
npx @omophub/omophub-mcp --api-key=oh_your_key --base-url=https://custom.api.com/v1

# HTTP mode
npx @omophub/omophub-mcp --transport=http --port=3100 --api-key=oh_your_key

# Stdio mode with standalone health endpoint
npx @omophub/omophub-mcp --api-key=oh_your_key --health-port=8080

Health Endpoint (Docker / Kubernetes)

In HTTP mode, the health endpoint is available at /health on the same port as the MCP endpoint:

npx @omophub/omophub-mcp --transport=http --port=3100 --api-key=oh_your_key
curl http://localhost:3100/health
# → {"status":"ok","version":"1.5.0","uptime_seconds":42}

In stdio mode, use --health-port for a standalone health endpoint:

HEALTH_PORT=8080 OMOPHUB_API_KEY=oh_your_key npx @omophub/omophub-mcp
curl http://localhost:8080/health

The Docker image defaults to HTTP mode on port 3100 with health checks built in.

Development

git clone https://github.com/OMOPHub/omophub-mcp.git
cd omophub-mcp
npm install
npm run build
npm test

Run locally:

OMOPHUB_API_KEY=oh_your_key npx tsx src/index.ts

Troubleshooting

Error	Solution
API key required	Set OMOPHUB_API_KEY in your environment or MCP config
Authentication failed	API key may be invalid or expired - generate a new one
Rate limit exceeded	Automatic retries are built in. For higher limits, upgrade your plan
Tools not appearing	Restart your AI client, verify npx @omophub/omophub-mcp runs without errors, check config path

Links

• Documentation
• Get an API Key
• Python SDK
• Community & Support

License

MIT - see LICENSE