Healthcare & Medical MCP Servers — FHIR, PubMed, Clinical Trials, DICOM, Drug Databases, and More

Healthcare and medical MCP servers are connecting AI agents to the full spectrum of clinical and biomedical data. Instead of navigating complex FHIR APIs, searching PubMed manually, or learning DICOM protocols, these servers let agents query patient records, search medical literature, look up drug interactions, browse clinical trials, and interact with imaging systems — all through the Model Context Protocol.

The landscape spans six areas: FHIR/EHR integration (multiple competing implementations for accessing electronic health records), medical research (PubMed and ClinicalTrials.gov access), multi-source healthcare hubs (servers bundling FDA, PubMed, ICD-10, and more), drug and pharmacology databases (DrugBank, RxNorm, NLM codes), medical imaging (DICOM/PACS interaction), and healthcare standards (HIPAA-compliant protocol extensions and public health data).

The headline findings: WSO2’s FHIR server pulls ahead at 120 stars (up from 110 tied with healthcare-mcp-public) — WSO2 leads EHR integration with SMART-on-FHIR auth, while healthcare-mcp-public bundles 9 tools across FDA, PubMed, clinical trials, ICD-10, and DICOM. PubMed-MCP-Server reaches 113 stars with deep paper analysis and PDF downloads, while cyanheads/pubmed-mcp-server offers 9 tools including citation generation, MeSH lookup, and a public hosted instance. DICOM-mcp reaches 95 stars as the leading imaging server with 11 tools for PACS interaction — and MedGemma 1.5’s new full DICOM support in Vertex AI opens a path to pairing MCP metadata retrieval with real image analysis. cyanheads/clinicaltrialsgov-mcp-server releases v2.4.2 (April 27, 2026) with continued architectural refinement. Infinitus launches Studio (April 23) — a no-code agent builder for prior auth and benefit verification workflows reporting 93% success rate, validating commercial healthcare MCP investment. HMCP adds specialty plugins for radiology and genomics. Medical-mcps unifies 14 biomedical APIs into 100+ tools through a single endpoint. Still no official Epic or Cerner MCP servers despite both expanding their FHIR API support. Keragon MCP remains in beta with 300+ healthcare integrations and SOC 2 Type II certification.

FHIR / EHR Integration

wso2/fhir-mcp-server (Most Popular FHIR Server)

WSO2’s open-source FHIR MCP server is now tied for the most starred server in the category. Its key value proposition: expose any FHIR server or API as an MCP server without custom coding. Provides full CRUD operations on FHIR resources:

• get_capabilities — retrieve metadata about available FHIR resource types
• search — execute FHIR search interactions with full parameter support
• read — retrieve individual resources by type and ID
• create / update / delete — full write operations on FHIR resources
• get_user — retrieve the authenticated user’s FHIR profile

Supports three transport modes (stdio, SSE, streamable HTTP) with SMART-on-FHIR authentication. Compatible with Claude Desktop, VS Code, and MCP Inspector. The Python package installs via pip, making setup straightforward.

jmandel/health-record-mcp (Innovative Query Interface)

Takes a fundamentally different approach to EHR access. Instead of exposing FHIR operations directly, it provides three powerful query tools that let agents explore clinical data flexibly:

• grep_record — text/regex search across all record data
• query_record — read-only SQL SELECT queries on FHIR data
• eval_record — custom JavaScript execution on fetched data

Uses SMART on FHIR standards for secure EHR connection via the SMART App Launch flow. Supports standalone web client, local stdio MCP server, and experimental SSE server modes. This design means agents can answer complex clinical questions without needing to know FHIR query syntax — they can write SQL or grep for patterns across the entire patient record.

the-momentum/fhir-mcp-server (Vector Search + FHIR)

The Momentum’s FHIR server combines traditional FHIR resource access with AI-powered document search. Beyond the 8 FHIR resource tools (patient, observation, condition, medication, immunization, encounter, allergy, generic), it adds:

• add_document_to_pinecone — ingest clinical documents into vector database
• search_pinecone — semantic search across stored documents using embeddings
• get_loinc_codes — medical observation standardization

This RAG-ready architecture means agents can semantically search clinical notes, discharge summaries, and other unstructured documents alongside structured FHIR data. Supports OAuth2 authentication, multi-format document ingestion (TXT, CSV, JSON, PDF), and Docker deployment.

AWS HealthLake MCP Server (Enterprise Cloud)

Part of AWS’s official MCP server collection, the HealthLake MCP server provides 11 tools for FHIR resource management with enterprise-grade features:

• Automatic datastore discovery — no manual configuration of HealthLake datastores
• Read-only mode — blocks all mutations for production/audit environments
• Advanced search — chained parameters, includes, revIncludes, modifiers, pagination
• FHIR job management — import and export operations
• Patient-everything — comprehensive patient data retrieval

Built with AWS SigV4 authentication, 235 tests with 96% coverage. The read-only mode is a thoughtful safety feature missing from most healthcare MCP servers — essential for deployments where AI agents should observe but not modify clinical data.

langcare/langcare-mcp-fhir (Enterprise EHR Proxy)

An enterprise-grade MCP server designed as a stateless proxy to any FHIR R4 EMR — Epic, Cerner, or GCP Healthcare API. Written entirely in Go for performance, with security features unusual in the MCP ecosystem:

• TLS 1.3 encryption — all communications encrypted in transit
• PHI scrubbing in logs — protected health information automatically removed from log output
• HIPAA-compliant audit logging — comprehensive activity tracking
• OAuth2 support — works with Epic, Cerner, and GCP authentication flows

Ships with a 40+ Clinical Skills Library — agent-agnostic workflow guides covering medication management, lab interpretation, clinical decision support, documentation, and population health. Also includes embedded React-based MCP Apps (FHIR Explorer, Patient Chart Review) that render directly in compatible clients.

New since March 2026: LangCare has added a Healthcare Voice Agent — real-time voice AI that lets patients ask about their health records and get spoken answers over FHIR data. Also ships a LangCare CLI — a Python wrapper that exposes FHIR tools (search, read, create, update) over HTTP for AI agent frameworks that don’t speak MCP natively (LangChain, smolagents, CrewAI, AutoGen). Stars have doubled from 15→31, signaling growing enterprise interest.

rkirkendall/medplum-mcp (Most Tools)

The most tool-rich FHIR server with 33 dedicated tools covering 8 FHIR resource types. Each resource gets full CRUD operations plus search — Patient, Practitioner, Organization, Encounter, Observation, MedicationRequest, Medication, and EpisodeOfCare. Also includes a general generalFhirSearch tool for querying any resource type. Designed specifically for the Medplum open-source healthcare platform.

Medical Research

JackKuo666/PubMed-MCP-Server (Most Popular Research Server)

The most starred medical research MCP server and a critical tool for biomedical literature access. Bridges AI assistants with PubMed’s database of more than 40 million citations:

• search_pubmed_key_words — keyword-based article search
• search_pubmed_advanced — advanced parameter-based queries
• get_pubmed_article_metadata — rapid metadata retrieval
• download_pubmed_pdf — full-text PDF download when available
• deep_paper_analysis — comprehensive paper analysis with structured output

The PDF download and deep analysis tools set this apart from basic PubMed search — agents can retrieve full papers and extract structured insights, not just titles and abstracts.

cyanheads/pubmed-mcp-server (NEW — Most Sophisticated PubMed Access)

A new competitor to JackKuo666’s PubMed server, offering 9 tools with significantly broader capabilities:

• pubmed_search_articles — query with boolean syntax, filters, and date ranges
• pubmed_fetch_articles — batch metadata retrieval (up to 200 PMIDs)
• pubmed_fetch_fulltext — PMC full-text extraction with section filtering
• pubmed_format_citations — generate APA 7th, MLA 9th, BibTeX, and RIS citations
• pubmed_find_related — discover similar, citing, and referenced articles via ELink
• pubmed_spell_check — NCBI ESpell biomedical query correction
• pubmed_lookup_mesh — MeSH vocabulary exploration with tree numbers
• pubmed_lookup_citation — ECitMatch partial reference resolution
• pubmed_convert_ids — DOI/PMID/PMCID conversion

Built on the @cyanheads/mcp-ts-core framework with Bun runtime, supporting both STDIO and Streamable HTTP transport. A publicly hosted instance is available at https://pubmed.caseyjhand.com/mcp — unusual for a research server. The citation generation, MeSH exploration, and ID conversion tools fill gaps that JackKuo666’s server doesn’t address, making this the more capable option for systematic reviews and reference management workflows.

cyanheads/clinicaltrialsgov-mcp-server (Best Clinical Trials Access)

The most sophisticated ClinicalTrials.gov MCP server, now at v2.4.2 (April 27, 2026) with continued architectural refinement. Goes well beyond basic search with analytical capabilities:

• clinicaltrials_search_studies — full-text and filtered search with geographic proximity
• clinicaltrials_get_study_record — comprehensive study information by NCT ID
• clinicaltrials_get_study_results — outcomes and adverse events for completed trials
• clinicaltrials_get_study_count — study count queries
• clinicaltrials_find_eligible — patient-to-trial matching with location-based sorting
• clinicaltrials_get_field_values — enum value discovery with study count frequency
• clinicaltrials_get_field_definitions — field schema discovery

The patient-to-trial matching tool is particularly valuable for clinical research coordinators — it can find eligible trials for a patient based on their conditions and location. The v2.x rewrite brings OpenTelemetry integration, dual transport (STDIO + Streamable HTTP), optional authentication, and a publicly hosted instance at https://clinicaltrials.caseyjhand.com/mcp. Active maintenance is evident from the v2.4.2 release (May 18, 2026) just days after the prior minor update.

pascalwhoop/medical-mcps (Most Comprehensive Biomedical)

The most ambitious medical MCP server, unifying 14 biomedical APIs into 100+ tools through a single endpoint:

• Literature: PubMed/PubTator3 for article search and annotation
• Drugs: OpenFDA for adverse events and drug labels, ChEMBL for compound data
• Genetics: MyVariant.info for variant annotation, GWAS Catalog for genome-wide associations
• Proteins: UniProt for protein data, MyGene.info for gene information
• Pathways: Reactome, KEGG, and Pathway Commons for biological pathway analysis
• Disease: OMIM and MyDisease.info for disease-gene relationships
• Clinical trials: ClinicalTrials.gov and NCI Clinical Trials APIs
• Knowledge graphs: Every Cure Matrix, OpenTargets Platform

Available as a unified cloud endpoint. This is a research powerhouse — a single MCP connection gives agents access to the major biomedical databases that researchers typically access through dozens of different interfaces.

JackKuo666/ClinicalTrials-MCP-Server

A Python-focused ClinicalTrials.gov server with CSV export capabilities — search trials, get full study details, generate statistics, and save results to CSV files for offline analysis. Also provides MCP resources for direct data access. A lighter alternative to the cyanheads server when you need data export more than analytical tools.

Multi-Source Healthcare Hubs

Cicatriiz/healthcare-mcp-public (Most Starred Healthcare Server)

Now tied with WSO2 as the most starred healthcare MCP server, providing a single entry point to multiple medical data sources:

• FDA Drug Information — search and retrieve drug data with multiple search types (general, label, adverse events)
• PubMed Research — medical literature search with date range filtering
• Health Topics — evidence-based health information from Health.gov (English/Spanish)
• Clinical Trials — search with status filtering and eligibility details
• Medical Terminology (ICD-10) — diagnosis code lookup by code or description
• medRxiv Search — pre-print biomedical articles
• NCBI Bookshelf Search — biomedical books and reference materials
• DICOM Metadata Extraction — parse metadata from medical imaging files
• Medical Calculator — BMI calculation

The breadth makes this ideal as a general-purpose medical assistant backend. Features caching, connection pooling, both HTTP/SSE and stdio interfaces, and Swagger UI documentation. 72 commits show steady development.

JamesANZ/medical-mcp (No API Keys Needed)

Differentiates with zero API key requirements — every data source is freely accessible, making setup trivial. Provides 15 tools across multiple medical domains:

• Drug tools — search drugs, get details, search nomenclature via FDA and RxNorm
• Literature — PubMed (30M+ articles), Google Scholar, multi-database search
• Statistics — WHO health statistics by country
• Pediatrics — AAP guidelines, pediatric drug search, child health statistics
• Clinical guidelines — evidence-based clinical guideline search

Runs 100% locally with no tracking, no logs, no cloud dependency. The pediatric focus is unique in the category — no other server provides dedicated AAP guideline and pediatric literature search tools.

Drug & Pharmacology

openpharma-org/drugbank-mcp-server

Access to 17,430+ pharmaceutical records via a high-performance SQLite backend. A single drugbank_info tool exposes 16 search methods:

• Search by name, indication, target, ATC code, structure, category, carrier, transporter, half-life
• Get drug details, interactions (drug-drug and food), pathways, products, external identifiers, salts, similar drugs

Covers 13,166 small molecule drugs plus 4,264 biotech compounds. The SQLite backend means fast queries with minimal memory overhead — no external API calls needed. Drug interaction lookup is particularly valuable for clinical decision support.

openpharma-org/nlm-codes-mcp

Provides access to clinical coding systems via the NLM Clinical Tables API. A single nlm_ct_codes tool supports 11 medical coding systems: ICD-10-CM, ICD-11, HCPCS Level II, NPI (organizations and individuals), HPO vocabulary, conditions, RxTerms, LOINC questions, NCBI genes, and major surgeries/implants. Essential for billing, clinical documentation, and regulatory compliance workflows.

johnyquest7/Medical_calculator_MCP

A focused MCP server for medical calculations — BMI, GFR, dosage calculations, and other clinical formulas. Integrates with Claude Desktop for clinical decision support.

Medical Imaging

ChristianHinge/dicom-mcp (Most Popular Imaging Server)

The leading medical imaging MCP server, enabling AI agents to interact with DICOM servers (PACS, VNA, etc.). Provides 11 tools across four categories:

Querying:

• query_patients — search by name, ID, birth date
• query_studies — find studies by patient ID, date, modality
• query_series — locate series within studies
• query_instances — find individual DICOM instances

Data Movement:

• move_series / move_study — send DICOM data to other configured nodes

Document Extraction:

• extract_pdf_text_from_dicom — extract text from PDF reports stored as DICOM objects

Utilities:

• list_dicom_nodes / switch_dicom_node — manage DICOM node connections
• verify_connection — test connectivity
• get_attribute_presets — retrieve query attribute configurations

The server requires a YAML configuration defining DICOM nodes and calling AE titles. This enables complex workflows like “find all MRI studies for patient X from the last 6 months, then move them to the research PACS” — tasks that normally require specialized DICOM tooling.

May 2026 update — MedGemma 1.5 bridges the analysis gap: Google’s MedGemma 1.5 (4B-parameter multimodal model) now includes full DICOM support in Vertex AI and can be used as an agentic tool with MCP function calling. While dicom-mcp retrieves and routes DICOM objects, MedGemma can analyze the actual image content — pairing the two starts to address the long-standing limitation of MCP imaging servers being metadata-only. This combination isn’t packaged as a single ready-to-deploy MCP server yet, but the integration path is clear.

fluxinc/dicom-mcp-server

A lighter DICOM server focused on connectivity testing — list nodes, C-ECHO by node name, and direct C-ECHO. Useful for verifying DICOM network configurations but lacks the query and data movement capabilities of ChristianHinge’s server.

Healthcare Standards & Public Health

innovaccer/Healthcare-MCP (HMCP Specification)

Not a tool server but a protocol specification — the Healthcare Model Context Protocol (HMCP) extends MCP with healthcare-specific capabilities:

• HIPAA-compliant security — access management with comprehensive audit trails
• OAuth 2.0 / SMART on FHIR — client credentials and authorization code with PKCE flows
• Multi-agent workflows — bidirectional agent-to-agent communication via sampling endpoints
• Patient context isolation — based on SMART on FHIR specifications

HMCP addresses the biggest gap in healthcare MCP adoption: there’s no standard way to ensure MCP servers handling PHI meet regulatory requirements. Innovaccer provides both the specification and an SDK (HMCPServer, HMCPClientConnector) for building compliant implementations. The specification is evolving with new specialty plugins for radiology and genomics workflows added to the specification — two domains previously identified as gaps in the broader healthcare MCP landscape. Real-time risk scoring for unsafe requests is also being added to the spec.

Infinitus Studio (Prior Auth Workflow Platform)

Infinitus launched its healthcare MCP server in late 2025 and followed up in April 2026 with Studio — a no-code agent builder for healthcare-specific workflows. Key commercial metrics from Studio’s launch:

• 93% success rate on benefit verification and prior authorization calls
• 40% greater accuracy than manually built agents
• 90% faster deployment from workflow design to production

The Infinitus platform specializes in the administrative workflows that consume the most clinical staff time — prior authorization, appeals follow-up, and benefit verification — connecting to payer systems via phone and data APIs. Unlike open-source servers that expose medical databases, Infinitus targets the payer-provider communication layer that most MCP servers ignore entirely. Represents the first major commercial validation of production-scale healthcare MCP workflows.

Keragon Healthcare MCP (HIPAA-Compliant Platform)

Keragon offers a commercial healthcare MCP platform purpose-built for regulated environments:

• 300+ healthcare integrations — EHRs, billing platforms, CRMs, scheduling systems
• Least-privilege access control — define exactly what data each AI agent can access, down to specific fields
• End-to-end audit logging — every tool call and payload logged for compliance
• HIPAA-compliant by design — SOC 2 Type II certified, built for regulated healthcare from the ground up

Still in beta with 100 free tool calls per day. Early beta results are encouraging — a multi-site outpatient clinic in Massachusetts reported double-digit decreases in administrative time using MCP-powered no-show reduction workflows. Software vendors building AI agents for healthcare can integrate Keragon MCP for free, instantly accessing all 300+ healthcare integrations without custom development. Unlike open-source alternatives, Keragon handles the compliance infrastructure so developers can focus on clinical workflows rather than security architecture.

Cicatriiz/pophive-mcp-server (Public Health Surveillance)

Provides access to Yale School of Public Health’s PopHIVE surveillance data — 11 datasets covering immunizations, respiratory diseases, chronic diseases, hospital capacity, injury/overdose tracking, and youth mental health. Featured on Claude. Tools include filter by state/demographics, compare states with statistical analysis, time series trend analysis, dataset catalog, and cross-dataset search. Production-ready for population health research.

What’s Missing

The healthcare MCP ecosystem has impressive breadth but notable gaps:

• No official EHR vendor servers — neither Epic nor Cerner has released an MCP server, despite both having extensive FHIR APIs. Community servers bridge the gap but lack vendor support.
• Medical imaging still limited to metadata in MCP — DICOM servers query and move imaging data but can’t analyze images natively. MedGemma 1.5’s new DICOM support in Vertex AI provides an integration path, but no packaged MCP server yet delivers end-to-end image retrieval + analysis.
• Genomics nearly absent — MyVariant.info is available through medical-mcps but there’s no dedicated genomics MCP server for 23andMe, gnomAD, or clinical sequencing data. HMCP’s new genomics specialty plugin may catalyze development here.
• Mental health gap — the only mental health MCP server (mcp-ai-therapy) is an AI therapy chatbot, not a clinical tool. No servers for PHQ-9 screening, therapy note management, or psychiatric assessment tools.
• HIPAA compliance unclear — most servers handle PHI but don’t document their compliance posture. Only LangCare, Keragon, and HMCP explicitly address HIPAA.
• No pharmacy workflow servers — drug databases exist but there’s nothing for e-prescribing, medication reconciliation, or pharmacy management systems.

The Bottom Line

Rating: 4.0 / 5

Healthcare MCP servers hold their strong rating thanks to impressive breadth — every major healthcare data domain has at least one MCP server, from EHR access to drug databases to medical imaging to public health surveillance. The FHIR subcategory has genuine competition (7+ implementations) driving quality and innovation. Medical research is the strongest subcategory: two competing PubMed servers (JackKuo666 at 113 stars, cyanheads at ~86 stars with 9 tools and a public hosted instance), plus the ClinicalTrials.gov server freshly updated to v2.4.2 (April 27, 2026) with OpenTelemetry and dual transport.

What elevates this category this cycle: Infinitus Studio demonstrates production-scale commercial healthcare MCP workflows with measurable clinical outcomes (93% success rate on prior auth), the strongest commercial validation yet of the platform’s real-world value. WSO2’s FHIR server pulls ahead to 120 stars. MedGemma 1.5’s DICOM integration in Vertex AI outlines a credible path to end-to-end imaging analysis paired with MCP retrieval. HMCP’s GxP profile trajectory and new radiology/genomics plugins signal the compliance layer maturing to meet the sector’s needs.

What holds it back from 4.5: still no official EHR vendor servers — Epic expanded FHIR API support for USCDI v5 but hasn’t released an MCP server. No packaged MCP server yet delivers actual medical image analysis (MedGemma + dicom-mcp pairing is an integration path, not a product). Compliance documentation remains spotty across most open-source projects. Specialized clinical domains (mental health, pharmacy workflows) remain underserved.

This review was last edited on 2026-05-20 using Claude Sonnet 4.6 (Anthropic).