Best MCP Servers for Healthcare in 2026

The best MCP servers for healthcare in November 2026 fall into two groups. The first is reference data: Medicare coverage rules, ICD-10 codes, the NPI Registry, PubMed and ClinicalTrials.gov, all offered as Anthropic-built connectors. The second is clinical data: FHIR (the standard format for health records) servers from AWS HealthLake, Medplum and CharmHealth, plus FDB's drug-knowledge server.

MCP (Model Context Protocol) is the open standard that lets an AI app such as Claude or ChatGPT read from and act in outside systems. In healthcare, that can mean "Is this procedure covered under the local LCD?" or "Pull this patient's active medications from our FHIR store." The second kind of question involves protected health information (PHI), and that changes everything about how you set it up.

We checked every entry against the vendor's own page on 2 November 2026. eSEOspace did not test these products. For the bigger picture of what MCP can and can't do in clinical settings, read our companion guide to MCP for healthcare.

How we chose

We used the MCP Guide's standard listicle criteria:

• Live and public. The vendor must publish a page or repository for its MCP server. We opened each one on 2 November 2026.
• Clear pricing. Prices come only from official pricing pages. Where none exists, we say "Pricing on request" or explain what the server costs to run.
• Security basics. The docs must describe sign-in, permissions, read-only modes or a compliance stance.

We dropped community FHIR and EHR wrappers that are not run by the data owner, and anything we could only see in news coverage. The list is short on purpose: healthcare has fewer credible, documented MCP servers than sales or DevOps. The order groups similar tools; it is not a ranking. Browse other lists on the MCP Guide's best-of hub.

Comparison at a glance

1. CMS Coverage: Medicare coverage rules in plain English

What it is. The CMS Coverage connector is "Made by Anthropic" and gives Claude access to "Medicare Part B coverage policies from the CMS Coverage Database, including National Coverage Determinations (NCDs) and Local Coverage Determinations (LCDs)." It has eight tools, including search_lcds, search_ncds and get_contractors.

Best for: Revenue cycle, prior authorization and appeals staff who look up coverage rules every day.

Pricing: No separate fee is listed. Anthropic's healthcare launch post (11 January 2026) says the new connectors are "generally available to all Claude subscribers."

Pros:

• No sign-in, and it only returns public policy data.
• Covers both national and local determinations.
• Remote server, so there is nothing to install.

Cons:

• Medicare Part B only, so it won't answer commercial payer questions.
• The listing warns it "may return data governed by third-party license agreements."

Checked against: Claude connector listing on 2 November 2026.

2. ICD-10 Codes: diagnosis and procedure lookups

What it is. The ICD-10 Codes connector is also made by Anthropic. It covers the ICD-10-CM (diagnosis) and ICD-10-PCS (procedure) code sets with nine tools, such as search_diagnosis_by_description, lookup_code and validate_code. Anthropic's launch post says the data comes from CMS and the CDC.

Best for: Medical coders, billers and anyone checking that a claim uses a valid code.

Pricing: No separate fee listed; it is part of the connectors Anthropic offers to Claude subscribers.

Pros:

• A validate_code tool catches invalid codes before submission.
• Search works by code or by plain-language description.
• No sign-in and no patient data needed.

Cons:

• It looks codes up; it does not decide which code a chart supports. A certified coder still has to.
• A different, unrelated "ICD-10" MCP server exists from another vendor, so check the URL (hcls.mcp.claude.com/icd10_codes/mcp).

Checked against: Claude connector listing on 2 November 2026.

3. NPI Registry: provider identity checks

What it is. The NPI Registry connector gives Claude access to the US National Provider Identifier Registry. It has three tools: npi_lookup, npi_search and npi_validate. Anthropic built it and lists it as added in January 2026.

Best for: Credentialing, provider directory and claims teams.

Pricing: No separate fee listed.

Pros:

• Small, focused tool set that is easy to approve.
• Useful for cleaning provider directories and referral lists.
• Public data, no sign-in.

Cons:

• The NPI Registry is not a license check. You still need state board verification for credentialing.
• Only three tools, so it is a helper rather than a workflow.

Checked against: Claude connector listing on 2 November 2026.

4. PubMed: biomedical literature with citations

What it is. Anthropic's PubMed connector "provides access to PubMed's biomedical citations and PubMed Central's full-text archive." Its tools include search_articles, find_related_articles, lookup_article_by_citation and get_full_text_article. The launch post says PubMed holds more than 35 million pieces of biomedical literature.

Best for: Clinicians, medical writers and researchers who need sourced answers.

Pricing: No separate fee listed.

Pros:

• Answers can carry real citations you can check.
• Full text where PubMed Central has it.
• A get_copyright_status tool helps before you reuse text.

Cons:

• Literature search is not clinical decision support. Evidence still needs expert reading.
• Full text is limited to what PubMed Central holds.

Checked against: Claude connector listing on 2 November 2026.

5. Clinical Trials: ClinicalTrials.gov for research teams

What it is. The Clinical Trials connector is Anthropic's server for ClinicalTrials.gov. Its six tools include search_trials, search_by_eligibility, search_by_sponsor, search_investigators and analyze_endpoints.

Best for: Research operations, trial matching and site feasibility work.

Pricing: No separate fee listed.

Pros:

• Eligibility search helps with patient-to-trial matching questions.
• Sponsor and investigator search support feasibility work.
• No sign-in.

Cons:

• Registry data can lag what sponsors actually do.
• Matching a real patient means sharing PHI with Claude, which brings your BAA setup into play.

Checked against: Claude connector listing on 2 November 2026.

6. AWS HealthLake MCP server: FHIR data on AWS

What it is. AWS Labs publishes an MCP server for AWS HealthLake, Amazon's managed FHIR data store. It offers 11 tools for datastore discovery, FHIR create, read, update and delete, search, patient_everything, and import and export jobs. You run it locally with uvx awslabs.healthlake-mcp-server@latest and your AWS credentials.

Best for: Health-tech and provider IT teams that already store FHIR data in HealthLake.

Pricing: The server is free and open source. HealthLake itself is billed by usage. The HealthLake pricing page lists $0.27 per data store hour, storage from $0.25 per GB-month (Standard) and query fees beyond 3,500 included queries per hour. The page also says "AWS HealthLake is a HIPAA-eligible service."

Pros:

• A --readonly flag blocks all five write and job tools.
• Uses normal AWS IAM, so access can be scoped and logged.
• HealthLake is HIPAA-eligible under your AWS agreement.

Cons:

• The sample IAM policy uses "Resource": "*"; tighten it before production.
• Without --readonly, the AI can delete FHIR resources.

Checked against: AWS Labs HealthLake MCP docs on 2 November 2026. Related profile: AWS MCP server.

7. Medplum: hosted MCP for a developer FHIR platform

What it is. Medplum is an open-source healthcare developer platform. Its MCP docs describe a hosted endpoint at https://api.medplum.com/mcp/stream with three tools. The main one, fhir-request, is "a powerful, low-level tool that performs a direct FHIR API request" using GET, POST, PUT, PATCH or DELETE.

Best for: Startups and digital health teams building their own apps on Medplum.

Pricing: The Medplum pricing page lists Free ($0, "for prototyping and learning"), Production from $2,000 per month, Premium from $6,000 per month and custom Enterprise pricing. It lists a "Standard BAA" under audit support and HITRUST e1 certification for the AWS-hosted platform; confirm which plan includes the BAA with sales.

Pros:

• OAuth sign-in, so users authorize with their own Medplum account.
• Full FHIR API reach from one tool.
• Open-source platform you can also self-host.

Cons:

• One broad tool can both read and modify data. There is no documented read-only switch on the MCP page.
• The Free plan is for prototypes, not patient care.

Checked against: Medplum MCP docs on 2 November 2026.

8. CharmHealth: an EHR vendor's own MCP server

What it is. CharmHealth publishes an MCP server for its EHR under its own GitHub organization. It has 16 tools for patients, encounters, medications, allergies, diagnoses, vitals, labs, appointments, messages and tasks. It runs over stdio by default, with an HTTP mode.

Best for: Practices and developers already on CharmHealth EHR.

Pricing: The server is free under the MIT license. API credentials come from CharmHealth API support, and we found no published price for API access, so treat it as pricing on request.

Pros:

• Built by the EHR vendor, not a third party.
• Covers the daily chart, scheduling and messaging work.
• A plain-language HIPAA notice explains the deployer's duties.

Cons:

• The README warns it "can access, transmit, and surface protected health information (PHI)" and should only be used with HIPAA-compliant LLM services under a signed BAA.
• Credentials are not self-service, and the HTTP mode needs a code change.

Checked against: CharmHealth GitHub repository on 2 November 2026.

9. FDB MedProof MCP: drug knowledge for AI agents

What it is. FDB (First Databank) calls MedProof MCP "the first Model Context Protocol server enabling safe agentic AI workflows with trusted Medication intelligence." It grounds AI agents in FDB's drug database, and FDB says it is generally available.

Best for: Health-tech companies building medication, refill or patient-messaging agents.

Pricing: Pricing on request through FDB's contact form.

Pros:

• Drug data from a long-standing medication knowledge vendor, not the open web.
• Aimed at reducing hallucinated drug facts.
• Works with any MCP-capable agent framework.

Cons:

• The product page does not list the tools, so you need a sales call to evaluate it.
• Built for software vendors, not for a single clinic to switch on.

Checked against: FDB MedProof MCP page on 2 November 2026.

HIPAA and BAA: the part most lists skip

A Business Associate Agreement (BAA) is the contract HIPAA requires before a vendor handles PHI for you. MCP adds a twist: data can flow from the AI app to a third server you did not sign anything with.

Anthropic's BAA article is clear on this. For Claude Enterprise, MCP connectors are "Available to use but sending data to 3rd parties via this feature isn't covered under Anthropic's BAA." It also says "Standard Claude Enterprise plans do not include BAA coverage without action from a Primary Owner," who must turn on HIPAA settings and accept the BAA.

In practice:

1. Keep PHI out of public-data connectors. You can ask about a code or a coverage rule without naming the patient.
2. Get a BAA with every system that holds PHI. That includes the FHIR store or EHR behind the MCP server, not just the AI vendor.
3. Start read-only. Use HealthLake's --readonly flag and limit roles in Medplum or CharmHealth.
4. Log every call. Our MCP audit logging guide covers what to record, and our piece on excessive permissions explains why broad roles are the biggest risk.

Run each candidate through the MCP server risk checklist before anyone connects real records.

What we left out

• Epic and athenahealth. We found community projects and news coverage, but no vendor-published MCP documentation we could open. athenahealth's own pages were not reachable when we checked.
• Community FHIR wrappers for Azure, Google Cloud and Epic sandboxes. They exist, but none is run by the platform owner.

Next step

If you need a healthcare MCP server with tighter scopes, audit logs and a design that keeps PHI where your BAAs cover it, our MCP development team builds and hardens servers for regulated data. Not sure you are ready? Start with an MCP readiness audit.

Frequently asked questions