FHIR R4
Any HL7 FHIR R4 server: search and read Patients, Observations, Conditions, Encounters, MedicationRequests, DiagnosticReports and any other resource type, plus the server's own capability statement. Bearer-token auth.
7-day free trial · No credit card · EU-hosted
Ask FHIR anything
Try a real prompt against FHIR R4 via AnythingMCP — Claude calls the tools, returns the answer. Press Enter on your own message to install on Cloud.
Install FHIR on CloudClaude is AI and can make mistakes. Please double-check responses.
Setup in 2 minutes
- 1Start a free trial on cloud.anythingmcp.com.
- 2Click FHIR R4 in the connector store. Paste credentials when prompted.
- 3Mint an MCP API key and point your AI client at the generated MCP URL.
Guides for FHIR
Step-by-step setup for the most popular AI clients.
Setting up FHIR R4
From the connector definition itself: where to find the credentials, what they need to be allowed to do, and which API this adapter talks to.
This is a generic FHIR R4 client, not a vendor connector. Point it at any R4 server — a hospital's HAPI instance, a national test server, a vendor's sandbox — and it speaks the standard.
- Set
FHIR_BASE_URLto the server's base, the URL that answers/metadata. No trailing slash. - Set
FHIR_ACCESS_TOKENto a bearer token. Most real deployments use SMART on FHIR, whose authorisation flow issues short-lived tokens; obtain one and paste it here, or use a long-lived service token if the server offers one. A server that needs no auth still needs a placeholder here — put anything.
Start with fhir_get_capability_statement. FHIR servers differ enormously in which resource types they hold and which search parameters they support, and /metadata is the machine-readable answer. Guessing produces a lot of confident 400s.
Search parameters are per resource type, and modifiers matter: name:contains=smith, birthdate=ge1980-01-01, code=http://loinc.org|85354-9, _lastUpdated=gt2026-09-01. Prefixes on numbers and dates are eq, ne, gt, lt, ge, le. A token search takes system|code, and the pipe is significant.
Results are a Bundle. The rows live in entry[].resource, the count in total (which some servers omit), and paging in link[] with relation: "next" — follow that link rather than computing an offset: pass its query string to fhir_next_page when it points at the server base (HAPI's ?_getpages=…), or to fhir_search with the same resource type when it points at a resource path. _count requests a page size; servers are free to ignore it.
_include and _revinclude fetch related resources in the same Bundle: _include=Observation:patient brings each observation's Patient along.
This is patient data. Anything read here is almost certainly health information under GDPR Article 9 or HIPAA. Whether an LLM may see it is a question about your legal basis and your processor agreements, not about this connector — decide it before installing, not after.
Read-only by design: no create, no update, no delete. A clinical record is not a thing an agent should write into by accident.
Your AI is two clicks away from FHIR R4.
Install the connector, paste your credentials, ask your AI a question. 7-day free trial, no credit card.