Try it free in the sandbox · first FHIR call in 5 minutesFree · 30 synthetic patients · no cardYour first API call in 2 minutes →

Connect ChatGPT and Claude to health data with MCP

Q: How do I connect Claude to health data?

Add https://platform.anpheros.com/mcp as a custom connector in Claude, sign in to Anpheros and approve. Claude then reads the parts of your record you allowed, read-only, and you can revoke access at any time.

How ChatGPT and Claude connect to health data through MCP connectors with OAuth and patient consent: Claude reads with permission, ChatGPT is write-only on real records, and what app developers need to build or not.

ChatGPT and Claude connect to outside data through MCP (Model Context Protocol) connectors: the assistant discovers your server, registers itself with OAuth, the person approves it, and the assistant then calls the tools you expose. For health data the hard part is not the protocol but consent: which assistant may read what, for how long, what it may write, and how the person sees and revokes it. Anpheros runs this as a service over a patient-controlled HL7 FHIR R4 record, so an app, a clinic or a person can connect an assistant without building an OAuth server, a consent screen and an audit trail.

Can ChatGPT read a patient's medical record? Through ChatGPT Health, adults in the US can connect records from participating providers. Through a custom MCP connector, ChatGPT can reach any server that implements MCP with OAuth. With Anpheros, ChatGPT is deliberately write-only on real records (it sends notes, symptoms and measurements as proposals the person accepts), and Claude reads the parts of the record the person allows, read-only.

At a glance

	Claude	ChatGPT	Your own agent
Connects through	Custom connector (MCP, OAuth)	Custom connector (MCP, OAuth)	MCP or REST/FHIR with a key or OAuth
On a real Anpheros record	Reads what the person allows, read-only by default	Write-only: notes, symptoms, measurements as proposals	Scopes your app requests and the person approves
On the sandbox	Full tools on 30 synthetic patients	Full tools on synthetic data	Full tools with an `sk_test_` key
Consent screen	Anpheros, names the assistant and what it may do	Same, with the OpenAI processing notice	Same
Revoke	In the Anpheros app (Connections) or in the assistant	Same	Same

Server address: https://platform.anpheros.com/mcp (Streamable HTTP). Discovery: https://platform.anpheros.com/.well-known/oauth-protected-resource/mcp.

How the connection works

The person adds https://platform.anpheros.com/mcp as a custom connector in the assistant.
The assistant reads the protected resource metadata (RFC 9728), finds the authorization server and registers itself with OAuth dynamic client registration (RFC 7591), using PKCE.
The person signs in to Anpheros and sees a consent screen that names the assistant, says that what it reads is processed by the company running it, and lists what it may do and for how long.
The assistant receives a token limited to that person's record and those scopes, and calls tools such as reading the timeline, reading results or noting a symptom.
Every call is logged; writes carry provenance (urn:anpheros:mcp); the person can revoke the connection at any time.

Why ChatGPT is write-only and Claude reads

Reading a whole medical history is a larger decision than saving a note. Our legal review chose to start ChatGPT with write-only access, so nothing in the record reaches ChatGPT through the connector, while still covering the most common need: people describe symptoms to ChatGPT and lose them when the chat ends. Claude reads with consent, read-only by default and section by section. Both rules live on the server; the assistant cannot widen them.

For app developers

If your app keeps health data in Anpheros, your users can connect an assistant to their own record without any work on your side: the consent, tokens, tool permissions and audit are handled by the platform. If you build your own agent, use the same MCP server with a sandbox key while developing and OAuth for real users. For ChatGPT apps that need to read data, design for the scopes a person would approve, label every result with its source and never put data in a URL.

Tools the server exposes

Reading: list patients (sandbox), AI-ready patient context, International Patient Summary, timeline, one measurement over time, compare two periods, search and read FHIR resources, provenance, LOINC lookup, health term search. Writing: note a symptom, a measurement or a short note in the person's record (as proposals on a real record), create a FHIR resource and import a lab report on the sandbox. The full list with parameters is in the MCP guide.

Frequently asked questions

How do I connect ChatGPT to medical records?

In the US, ChatGPT Health connects records from participating providers for adults. Anywhere, ChatGPT can use a custom MCP connector: with Anpheros it sends notes, symptoms and measurements to the person's record as proposals, without reading the record.

How do I connect Claude to health data?

Add https://platform.anpheros.com/mcp as a custom connector in Claude, sign in to Anpheros and approve. Claude then reads the parts of your record you allowed, read-only, and you can revoke access at any time.

Is MCP secure enough for health data?

MCP is a transport for tools. Security comes from the server: OAuth with PKCE, scopes the person approves, a consent screen that names the assistant, server-side tool allowlists, an audit log and provenance on writes. Those are what Anpheros adds.

Do I need my own OAuth server to build a health connector?

Not if the data lives in Anpheros: the platform runs discovery, dynamic client registration, consent and tokens. If you build your own MCP server, you need all of these.

Can I test with ChatGPT or Claude without real data?

Yes. The sandbox has 30 synthetic patients; connect with a sandbox key in Claude Code, Cursor or VS Code, or through OAuth on sandbox records.

Ask your AI

Paste this into Claude, ChatGPT, Cursor or Claude Code. Your assistant reads this guide and adapts it to your code.

Read https://developers.anpheros.com/guides/chatgpt-claude-health-connectors.md and help me apply this guide to my own project. Use the Anpheros sandbox (base URL https://platform.anpheros.com) with my API key from the ANPHEROS_KEY environment variable, never written in the code. Give me complete, runnable code, the response I should expect, and how to fix the errors in https://developers.anpheros.com/guides/troubleshooting.md if a call fails.

Open in Claude Open in ChatGPT This guide as Markdown Build the whole integration with an AI coding tool