Clinical MCP Servers

ColabBio is the first medical AI platform to implement standardized clinical microservices over the Model Context Protocol (MCP) via FastMCP and Server-Sent Events (SSE).

graph LR
    subgraph Agent ["LangGraph4j Clinical Companion"]
        CORE["Clinical Reasoning Core"]
    end

    subgraph MCPStack ["Clinical MCP Microservices"]
        FHIR["mcp-fhir<br/>(EHR Resources)"]
        OMERO["mcp-omero<br/>(WSI Slides & ROIs)"]
        OMOP["mcp-omop<br/>(OHDSI CDM & SNOMED)"]
        DICOM["mcp-dicom<br/>(PACS QIDO/WADO)"]
    end

    CORE <-->|JSON-RPC 2.0 / SSE| FHIR
    CORE <-->|JSON-RPC 2.0 / SSE| OMERO
    CORE <-->|JSON-RPC 2.0 / SSE| OMOP
    CORE <-->|JSON-RPC 2.0 / SSE| DICOM

---

The 4 Clinical MCP Servers

1. mcp-omop (Observational Medical Outcomes Partnership)

• Protocol: FastMCP over SSE (Port 8080).
• Capabilities:
  • Cohort query execution on Postgres/OMOP CDM.
  • Concept translation and vocabulary lookup (SNOMED-CT, RxNorm, LOINC).
  • Population-level patient baseline matching.

2. mcp-dicom (DICOMweb & PACS Archive)

• Protocol: FastMCP over SSE (Port 8081).
• Capabilities:
  • QIDO-RS: Query radiological studies by patient ID, accession number, and modality (CT, MR, MG, US).
  • WADO-RS: Resolve series streaming URIs and MedVFS virtual mount paths.
  • Extraction of Structured Reports (SR) and DICOM metadata.

3. mcp-fhir (Fast Healthcare Interoperability Resources)

• Capabilities:
  • Fetch patient demographics, condition histories, genomic diagnostic reports, and medication lists (R4).
  • Structured resource mapping with strict privacy guards.

4. mcp-omero (Whole Slide Imaging Repository)

• Capabilities:
  • Query pathology datasets, biopsy slides, and stain types (H&E, IHC, Ki-67).
  • Resolve gigapixel tile servers and spatial ROI annotations.

---

Deployment with Helm

The clinical MCP stack is deployed as modular microservices via Helm:

helm install colabbio-mcp ./cb-devops/cluster/charts/mcp-stack \
  --namespace colabbio-dev