FHIR Clinical Data Connector
Connect to a hospital FHIR R4 server and extract structured findings from a patient’s clinical data and notes.
The skill discloses the FHIR-server, clinical-note, and temporary-file data flows, forbids implicit connection, treats document content as untrusted, and requires user participation in SMART login. However, the default user/*.rs scope is broad, static bearer tokens are supported, and explicit per-read confirmation, credential-protection details, and verifiable rollback are absent, so points are deducted.
The connection, patient-matching, document-extraction, failure-continuation, and disconnect procedures are fairly coherent and include abnormal-input handling. However, the skill depends on multiple MCP tools, another skill, and a script, with no skill-specific tests or static reproducibility evidence; the instruction to improvise can also produce variable behavior. The static ceiling therefore applies.
Trigger phrases, target scenarios, and principal inputs and outputs are reasonably clear, covering several EHR vendors and SMART-on-FHIR. Non-fit boundaries, Chinese-language interaction, and mainland-China network reachability are not addressed, and production setup depends on user IT information, so points are deducted.
The documentation is layered around connection, retrieval, content fetching, extraction, and disconnect, with prerequisites, parameter examples, and some limitations. Versioning, changelog, maintenance ownership, troubleshooting, and license details are incomplete; the README supplies only a general terms statement, so points are deducted.
The core workflow can retrieve FHIR document text and pass it to structured extraction, with a defined output direction and fallback handling for binary documents. However, no representative outputs, correctness metrics, or comparative-benefit evidence are provided, and results still require human review, so the static ceiling applies and points are deducted.
The source gives an auditable procedure, named tools, and several constraints, but contains no committed tests covering key paths, CI execution evidence, or independent third-party corroboration. The assessment therefore relies mainly on documented claims and receives a low score.
- The default user/*.rs scope may allow cross-patient access; deployments should prefer a single-patient scope and confirm patient matching and data retrieval.
- Clinical data and SMART credentials are highly sensitive; logging, caching, inter-process transfer, and credential-storage behavior are not specified.
- Reachability of FHIR endpoints, browser login, and external sandboxes is not established; mainland-China users should verify access to the hospital endpoint and related services.
- No key-path tests or representative extraction results are supplied; structured outputs should be reviewed by qualified clinical personnel before use.
What it does & when to use it
This skill connects to Epic, Oracle Health/Cerner, MEDITECH, athenahealth, or another SMART-on-FHIR endpoint. It uses a local FHIR MCP server to find patients, retrieve clinical resources and DocumentReference content, and pass note text to clinical-note-extract for structured extraction. It is designed for questions that start from a live EHR, not text that has already been pasted into the conversation. Adoption requires a configured FHIR service and appropriate SMART login access.
Checks and connects to the FHIR R4 MCP server; searches for and confirms patients by name, date of birth, or MRN; reads Conditions, Observations, MedicationRequests, Allergies, DocumentReferences, and other FHIR resources; filters data by date range; retrieves note content and, when needed, routes PDF, DOCX, or scanned documents through doc-extract; sends untrusted note text and the user's question to clinical-note-extract; disconnects when the work is finished.
- A clinician needs to find a patient in a hospital EHR and review records from a specified time window.
- A researcher needs to summarize the last six months of clinical notes and extract structured findings about a defined question.
- A healthcare team using Epic, Cerner, or another SMART-on-FHIR system needs to read patient data from a live FHIR R4 endpoint.
- A healthcare data engineer needs to retrieve Encounter, Procedure, DiagnosticReport, Coverage, or other FHIR resources.
Pros & cons
- Supports several widely used EHR vendors plus generic SMART-on-FHIR endpoints.
- Can retrieve multiple FHIR resource types and apply date-range filters.
- Handles text, HTML, RTF, and XML/C-CDA narrative attachments.
- Delegates clinical-note structuring to a dedicated extraction skill while retaining document identifiers.
- Requires a functioning local FHIR MCP server, network access, and suitable SMART permissions.
- Production hospital endpoints require organization- or vendor-provided URL and client_id details.
- Binary documents may require the additional doc-extract workflow; the source provides no test-suite or vendor-validation evidence.
- The skill itself does not write to disk, but document extraction uses temporary files.
How to install
Add the repository as a plugin marketplace source and install the healthcare plugin: /plugin marketplace add anthropics/healthcare, then run /plugin install healthcare@healthcare. The FHIR skill is located at plugins/healthcare/skills/fhir/SKILL.md. The source does not document another installation method or a version requirement.
How to use
After installing the plugin and configuring an accessible FHIR R4 service, start with a request such as “connect to Epic” or “pull notes for patient X.” If FHIR_BASE_URL is not configured, specify the EHR or sandbox first; the source gives https://launch.smarthealthit.org/v/r4/fhir as an open SMART Health IT sandbox example. Production endpoints and secured sandboxes require the organization's FHIR base URL and registered SMART client_id. The skill does not connect implicitly on first use, and authentication may require browser-based sign-in.
Compared to similar skills
The source material provides no explicit alternative product or competitor comparison.