Pharmaceutical and healthcare MCP servers are connecting AI agents to electronic health records, biomedical literature, drug databases, medical imaging systems, genomic data, and clinical trial registries. Instead of navigating complex FHIR APIs or manually searching PubMed, an AI agent can query a patient’s EHR through SMART on FHIR authentication, search 36 million biomedical articles, analyze drug interactions across 17,000+ compounds, or retrieve DICOM imaging data from a hospital PACS — all through standardized MCP tools.
The landscape divides into seven areas: EHR/FHIR integration (WSO2, health-record-mcp, FHIR-MCP — bridging AI to electronic health records), healthcare platforms (Innovaccer HMCP extending the protocol itself, healthcare-mcp-public as a general medical data gateway), drug discovery (ChEMBL for compound research, DrugBank for pharmaceutical data), medical literature (5+ PubMed implementations, medical-mcp for multi-source queries), medical imaging (dicom-mcp for PACS/VNA interaction), genomics and bioinformatics (NCBI Datasets, MCPmed), and the OpenPharma initiative (45 coordinated repositories covering FDA, EMA, clinical trials, and more).
The headline findings: this is the deepest vertical MCP category we have reviewed — not just by server count but by maturity, institutional backing, and protocol-level innovation. FHIR has proven to be the perfect complement to MCP — healthcare already had a standardized data format, so MCP servers could focus on the AI integration layer rather than data normalization. Security is taken seriously — multiple servers implement HIPAA-compliant audit logging, PHI protection, and SMART on FHIR authentication, reflecting the regulatory reality of healthcare data. The OpenPharma initiative is unprecedented — 45 coordinated MCP servers for pharmaceutical data is the largest organized effort in any vertical. Drug discovery and biomedical research are the strongest use cases — ChEMBL’s 22 tools and multiple PubMed implementations show genuine demand from researchers.
EHR & FHIR Integration
WSO2 fhir-mcp-server
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| wso2/fhir-mcp-server | ~98 | Python | Apache 2.0 | 7 tools |
The most polished FHIR-to-MCP bridge available, from enterprise integration company WSO2. Exposes any FHIR-compliant server as an MCP server with:
- SMART on FHIR authentication — OAuth 2.0 Authorization Code Grant flow for secure EHR connections
- Multi-transport support — stdio, SSE, and streamable HTTP protocols
- Full CRUD operations — get_capabilities, search, read, create, update, delete, plus get_user
- Docker deployment — Docker Compose for easy setup
- Client compatibility — tested with VS Code, Claude Desktop, and MCP Inspector
The value here is abstraction — rather than building a custom integration for each EHR system, WSO2’s server turns any FHIR-compliant backend into an MCP-compatible data source. Given WSO2’s enterprise pedigree (they build integration middleware used by hundreds of organizations), this has more production credibility than most community projects.
health-record-mcp
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| jmandel/health-record-mcp | ~75 | TypeScript | MIT | 3 tools |
A secure gateway enabling AI to access patient data from Epic and Cerner EHRs, created by Josh Mandel — notably, a co-architect of the SMART on FHIR standard itself. Three powerful analytical tools:
- grep_record — text and regex search across all record data
- query_record — SQL SELECT queries on FHIR data
- eval_record — custom JavaScript execution on EHR data
The combination of SMART on FHIR authentication with flexible query tools (grep, SQL, JavaScript) gives this server unusual analytical depth. The standalone web client handles EHR OAuth connections, while the MCP server provides the AI interface. Supports both local stdio mode and full SSE-based server with OAuth 2.1. The creator’s deep FHIR expertise makes this one of the most architecturally sound healthcare MCP implementations.
FHIR-MCP
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| xSoVx/fhir-mcp | ~0 | TypeScript | MIT | 8 tools |
The most security-focused healthcare MCP server, with enterprise-grade hardening that goes well beyond typical open source projects:
- OWASP-compliant security — security headers, input validation, SQL injection prevention
- ML-powered PHI classification — dynamic masking and redaction of protected health information
- Break-glass emergency access — mechanism for authorized emergency data access
- Multi-tier rate limiting — PHI-aware throttling with different limits for sensitive vs. non-sensitive operations
- HIPAA-compliant audit logging — FHIR AuditEvent standards, comprehensive trails with redacted logging
- Role-based access control — SMART on FHIR/OAuth2 support
- Real-time suspicious activity detection — monitoring for anomalous access patterns
Tools cover 5 FHIR operations (capabilities, search, read, create, update) and 3 HL7 terminology services (lookup, expand, translate). Despite zero stars, the security implementation is genuinely impressive — this is what healthcare MCP servers should look like from a compliance perspective. The “not meant for clinical use” disclaimer is responsible but undersells the engineering quality.
AgenticCare
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| yatstan/AgenticCare | ~0 | JavaScript/TypeScript | MIT | 16 tools |
An MCP server for Epic and Cerner EMR interaction combining 13 FHIR tools with 3 medical research tools. Features OAuth2 authentication, response caching, and Claude Desktop integration. Provides a practical bridge for healthcare professionals who want AI-powered analysis of patient data across two of the largest EHR platforms in the US market.
Healthcare Platforms & Standards
Innovaccer HMCP (Healthcare Model Context Protocol)
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| innovaccer/Healthcare-MCP | ~28 | Python | MIT | Protocol extension |
The most ambitious structural contribution to healthcare MCP — rather than building another server, Innovaccer extended the MCP protocol itself with healthcare-specific capabilities:
- Patient context isolation — based on SMART on FHIR specifications, ensuring proper context boundaries
- SMART on FHIR OAuth — client credentials and authorization code flows
- Bidirectional agent communication — sampling capability enabling HMCP client and server to both communicate using LLM inputs/outputs
- HIPAA-compliant security — data segregation, encryption, comprehensive audit logging
- Multi-transport — SSE and streamable HTTP protocols
- Low-code agent builder — no-code interface for building additional healthcare AI agents
HMCP is important not just as a tool but as a signal — when a major healthcare data platform (Innovaccer processes data for health systems covering 55+ million lives) invests in extending MCP for healthcare, it validates the protocol’s relevance in regulated industries. The specification is open (MIT licensed), inviting the broader healthcare AI community to build on it.
healthcare-mcp-public
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| Cicatriiz/healthcare-mcp-public | ~102 | Node.js | — | 9 tools |
The most popular general-purpose medical MCP server — a one-stop shop for medical data access:
- FDA drug lookup — drug information from the FDA database
- PubMed search — biomedical literature queries
- Health topics — general health information
- Clinical trials search — ClinicalTrials.gov integration
- ICD-10 code lookup — diagnostic code queries
- medRxiv search — preprint access
- Medical calculator — BMI and basic clinical calculations
- NCBI Bookshelf search — medical reference text access
- DICOM metadata extraction — medical imaging metadata
Features efficient caching with connection pooling, both stdio and HTTP/SSE interfaces, and Swagger UI documentation. The breadth of coverage makes this ideal for developers who want a single MCP server that covers the major medical data sources without configuring multiple specialized servers.
medical-mcp
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| JamesANZ/medical-mcp | ~75 | TypeScript | MIT | 15 tools |
Zero-configuration medical data access — queries FDA, WHO, PubMed, RxNorm, and Google Scholar with no API keys required. Key selling points:
- Local-only operation — no data leaves your machine (important for medical privacy)
- No API keys — uses public endpoints only
- 15 tools — drug search and details, WHO health statistics, PubMed literature search, Google Scholar academic research, clinical guidelines, pediatric-specific resources
- In-memory caching — TTL-based cache with statistics monitoring
The privacy-first design makes this particularly suitable for medical professionals who want AI assistance without sending patient-adjacent queries to third-party services. At 75 stars, it has genuine community adoption.
Clinical Medication Intelligence
FDB MedProof MCP (First Databank)
| Server | Type | License | GA |
|---|---|---|---|
| FDB MedProof MCP™ | Commercial (enterprise) | Proprietary | March 31, 2026 |
The first MCP server built specifically for AI agent-driven medication decisions. First Databank — whose drug databases are already embedded in most major EHRs — launched MedProof MCP to bring that same medication intelligence to AI agents operating across clinical workflows.
MedProof MCP connects AI agents to FDB’s drug knowledge database across major EHR platforms including Epic, athenahealth, eClinicalWorks, MEDITECH, and Oracle Health (Cerner). This means agents can perform drug interaction checks, dosage validation, and medication decision support in patient-specific context without requiring custom integrations per EHR.
FDB also announced two complementary AI solutions:
- FDB Script Agent™ — a prescription automation agent for ambulatory settings where physicians dictate prescriptions that get structured and queued for review
- FDB VerifyAssist™ — an inpatient pharmacy verification assistant that runs drug checks at the point of need, reducing chart-digging time for pharmacists
The platform is already in production at Artera — a patient communications platform serving over 1,000 healthcare organizations and more than 100 million patients annually. This makes FDB MedProof MCP one of the largest-scale healthcare MCP deployments documented to date.
This is a notable shift: a major clinical data vendor building MCP-native products rather than retrofitting existing APIs. It signals that MCP is becoming the expected integration standard for healthcare AI, not just an option.
Drug Discovery & Pharmaceutical Data
ChEMBL-MCP-Server
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| Augmented-Nature/ChEMBL-MCP-Server | ~77 | TypeScript | — | 22 tools |
The most comprehensive drug discovery MCP server, providing advanced access to the ChEMBL chemical database through 22 specialized tools across five categories:
- Core chemical search (5 tools) — compound lookup, structure search, similarity matching
- Target analysis (5 tools) — protein target identification, mechanism of action
- Bioactivity data (5 tools) — assay results, potency measurements, selectivity profiles
- Drug development (4 tools) — clinical pipeline tracking, approval status, indication mapping
- Chemical properties (4 tools) — ADMET analysis, physicochemical properties, advanced filtering
ChEMBL contains bioactivity data for 2.4 million compounds against 15,000+ targets — having 22 MCP tools to navigate this database transforms drug discovery research workflows. A medicinal chemist using Claude Desktop could search for compounds active against a specific target, filter by potency and selectivity, check ADMET properties, and review the clinical pipeline — all in a single conversation.
DrugBank MCP Server
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| openpharma-org/drugbank-mcp-server | ~2 | JavaScript | MIT | 1 tool (16 methods) |
Access to 17,430+ drugs (13,166 small molecules + 4,264 biotech) via a high-performance SQLite backend with sub-10ms query speeds and ~50–100MB memory usage:
- Drug name and indication searching
- Target protein, enzyme, carrier, and transporter identification
- Drug-drug interaction detection
- Metabolic pathway analysis
- Chemical structure queries (SMILES/InChI)
- Half-life range filtering and drug similarity scoring
- Market product information and external database cross-referencing
The SQLite approach means the entire DrugBank dataset runs locally without API calls — fast, private, and offline-capable. Despite low stars, the data coverage is genuinely useful for pharmaceutical research workflows.
Medical Literature & Research
PubMed MCP Servers
PubMed is the most replicated MCP server category in healthcare — at least 5 independent implementations exist, reflecting strong demand from biomedical researchers:
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| cyanheads/pubmed-mcp-server | ~66 | TypeScript | Apache 2.0 | 7 tools |
| Augmented-Nature/PubMed-MCP-Server | — | TypeScript | — | 16 tools |
| JackKuo666/PubMed-MCP-Server | — | Python | — | Multiple |
| chrismannina/pubmed-mcp | — | TypeScript | — | Multiple |
| openpharma-org/pubmed-mcp | — | JavaScript | — | Literature search |
The cyanheads implementation leads with 66 stars, 7 tools, full NCBI E-utilities integration (ESearch, EFetch, ESummary, ELink, ESpell, EInfo), citation formatting in APA 7th/MLA 9th/BibTeX/RIS, full-text article retrieval from PubMed Central with section filtering, and Cloudflare Workers deployment. The Augmented-Nature version offers 16 tools exposing the full NCBI E-utilities and PMC APIs.
The proliferation of PubMed MCP servers tells a clear story: biomedical literature search is the killer app for healthcare MCP. Researchers want to query 36 million citations from within their AI assistant rather than context-switching to the PubMed website.
NCBI MCP Server
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| vitorpavinato/ncbi-mcp-server | — | Python | — | Multiple |
Designed for researchers across life sciences, providing access to PubMed’s 35+ million articles through natural language queries. Focused on making NCBI’s E-utilities accessible without requiring researchers to learn the API syntax.
Medical Imaging
dicom-mcp
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| ChristianHinge/dicom-mcp | ~86 | Python | MIT | 10 tools |
The only MCP server for medical imaging interoperability, enabling AI assistants to interact with PACS (Picture Archiving and Communication Systems) and VNA (Vendor Neutral Archives):
- Query metadata (4 tools) — search patients, studies, series, and instances by name, ID, date, modality
- Read reports (1 tool) — extract text from encapsulated PDF reports in DICOM format
- Send images (2 tools) — route imaging data to other DICOM destinations
- Utilities (3 tools) — connection management and network verification
At 86 stars, this fills a critical niche — radiology departments generate massive amounts of imaging data, and being able to query study metadata, read reports, and route images through an AI assistant could streamline radiologist workflows significantly. The disclaimer that it’s “not meant for clinical use” is appropriate given the sensitivity of medical imaging data, but the technical implementation is solid.
Genomics & Bioinformatics
NCBI-Datasets-MCP-Server
| Server | Stars | Language | License | Tools |
|---|---|---|---|---|
| Augmented-Nature/NCBI-Datasets-MCP-Server | ~11 | TypeScript | MIT | 31 tools |
The most comprehensive genomics MCP server, providing access to NCBI’s datasets through 31 specialized tools across nine categories:
- Genome data retrieval and analysis
- Gene information and search
- Taxonomy classification
- Assembly data and quality metrics
- Virus genome data
- Protein sequence analysis
- Genome annotation
- Comparative genomics
- Sequence and phylogenetic analysis
With resource templates for direct URI-based data access, rate limiting, and caching, this server makes NCBI’s vast genomic databases accessible to AI-assisted bioinformatics workflows. A researcher could ask their AI assistant to find all known variants of a gene, compare assemblies across species, or retrieve protein sequences — all through natural language.
Additional Bioinformatics Servers
- bio-mcp — protein structure analysis capabilities for LLMs
- MCPmed GEO server — Gene Expression Omnibus data access via NCBI E-Utils
- ClinVar MCP (openpharma-org) — variant interpretation via NCBI E-utilities
- GWAS Catalog MCP (openpharma-org) — genome-wide association study data
- COSMIC MCP (openpharma-org) — somatic mutations in cancer data
The OpenPharma Initiative
The openpharma-org GitHub organization maintains 45 repositories — the largest coordinated MCP server collection for any industry vertical. Key servers include:
| Server | Focus | Language |
|---|---|---|
| fda-mcp | FDA drug labels, adverse events, recalls, safety data | TypeScript |
| ema-mcp | European Medicines Agency approvals, EPARs, orphan designations | JavaScript |
| drugbank-mcp-server | 17,430+ drug database | JavaScript |
| clinicaltrials-mcp | ClinicalTrials.gov API v2 | — |
| pubmed-mcp | Biomedical literature search | JavaScript |
| cdc-mcp | CDC disease surveillance, public health data | TypeScript |
| nlm-codes-mcp | ICD-10/11, HCPCS, NPI, HPO medical codes | JavaScript |
| patents-mcp | USPTO and Google Patents search | Python |
| medicare-mcp | CMS Medicare provider and claims data | JavaScript |
| biorxiv-mcp | 260K+ bioRxiv/medRxiv preprints | JavaScript |
| hmdb-mcp-server | Human Metabolome Database | JavaScript |
| cosmic-mcp-server | Catalogue of Somatic Mutations in Cancer | JavaScript |
| clinvar-mcp-server | ClinVar variant interpretation | JavaScript |
| gwas-mcp-server | NHGRI-EBI GWAS Catalog | JavaScript |
| sec-mcp | SEC EDGAR filings (pharma company financials) | JavaScript |
All servers use public APIs only — no proprietary databases or subscriptions required. MIT licensed. The initiative’s scope is remarkable: a pharmaceutical researcher could connect all these servers and have AI-powered access to drug data from both FDA and EMA, clinical trial results, biomedical literature, patent landscapes, genomic variants, metabolomics data, and financial filings — a comprehensive pharmaceutical intelligence toolkit.
Notable Gaps
Despite being the deepest vertical MCP category, some areas remain uncovered:
- Pharmacy dispensing — FDB MedProof MCP covers medication decision support and prescription automation (Script Agent), but pharmacy dispensing workflow integration (inventory, fill tracking, patient pickup) remains uncovered
- Clinical decision support — FDB MedProof MCP addresses medication CDS specifically, but no general-purpose rule-engine-based CDS beyond what EHR systems provide internally
- Insurance claims adjudication — no provider-side claims processing
- Clinical documentation — no nursing notes, physician documentation, or medical transcription MCP servers
- IoMT (Internet of Medical Things) — beyond DICOM, no medical device integration (vital signs monitors, infusion pumps, wearables)
- Population health analytics — no cohort analysis or epidemiological modeling tools
- Quality measures — no eCQM/HEDIS reporting
- Surgical workflows — no OR scheduling, surgical planning, or procedure tracking
- Pathology/Lab systems — no LIS (Laboratory Information System) integration
- EMS/emergency dispatch — no ambulance or emergency medical services coordination
The Bottom Line
Pharmaceutical and healthcare represents the gold standard for vertical MCP development. The combination of genuine depth (40+ servers across 7 subcategories), production-grade security (HIPAA-compliant audit logging, PHI protection, SMART on FHIR authentication), protocol-level innovation (Innovaccer’s HMCP), institutional backing (WSO2, Innovaccer, OpenPharma), and the largest coordinated server collection of any industry we have reviewed makes this category exceptional.
The key enabler is FHIR — healthcare already had a standardized data interchange format, which meant MCP servers could focus on the AI integration layer rather than solving data normalization problems. Other verticals could learn from this: standardize your data format first, and the AI integration layer becomes dramatically easier to build.
Rating: 4.5/5 — The highest-rated vertical MCP category. Depth, maturity, security consciousness, and institutional investment are all exceptional. The 0.5 deduction reflects the remaining gaps in clinical workflow integration (pharmacy, documentation, population health) and the fact that many individual servers are still early-stage despite the category’s overall maturity.
See Also
- Utah Lets AI Renew Prescriptions Without a Doctor — the first state-approved autonomous AI prescribing program, the jailbreak that exposed its vulnerabilities, and the expansion into psychiatric medications
This review was last edited on 2026-04-04 using Claude Opus 4.6 (Anthropic).