Open standards
The Model Context Protocol
MCP is an open standard for connecting AI assistants to real systems. Max Health builds on it so you can point an assistant at your own health data and ask questions of it, in whichever client you already use.
What MCP is
Before MCP, connecting an assistant to a system meant a bespoke integration for that one pairing. Every new client had to redo the work, and every system had to court every client. MCP replaces that with one contract: a system runs an MCP server, and any MCP-capable client can speak to it.
A server offers three kinds of thing. Tools are actions the model may call. Resources are documents or data it may read. Prompts are prepared instructions a client can offer the user. That is the whole surface, which is why support arrived quickly across clients.
Why it matters for health data
Health data is the hard case for AI, and not because the models are weak. It is scattered across portals, clinics, labs and devices; it arrives as PDFs, faxes and FHIR bundles; and the questions worth asking span all of it at once. “Has this marker moved since 2019” is trivial to ask and, without a connection to your records, impossible to answer.
MCP is a good fit precisely because it is boring plumbing: it standardises the connection and leaves the data where it is. Nothing has to be copied into a vendor's silo for an assistant to read it, and because the protocol is open, the same endpoint serves whichever client you prefer. That is the same reason Max Health builds on FHIR and other open standards everywhere else.
Connectors, skills and plugins
These four words get used interchangeably in the wild. They mean different things, and the distinction matters when you are deciding what to install.
- Connector
- One MCP endpoint.
- Skill
- A task-scoped instruction package — a SKILL.md a client can load.
- Plugin
- Connectors and skills bundled together, installable as one unit.
- MCP Apps
- The extension for returning interactive ui:// resources instead of raw JSON.
Max Health MCP endpoints
6 of the 6 plugins consumable by an agent publish a hosted endpoint; the rest run inside your own infrastructure. Endpoints that reach account-scoped data require you to sign in first, and show you only what that account can already see.
AIHR
The AI-native health record: records access, clinical AI, and the parametric 3D body.
Endpoint https://aihr.maxhealth.tech/mcp
Skills: patient-chart-review, clinical-ordering
DICOM Viewer
Browser-native 3D medical image viewer, zero install.
Endpoint https://dicom.maxhealth.tech/mcp
Skills: imaging-study-review
Legal Research Tools for AI
Multi-jurisdiction legal research and the AI patient-rights advocate.
Endpoint https://legal.maxhealth.tech/mcp
Skills: patient-rights-advocate, legal-research
Trust
Document signing and trust services.
Endpoint https://trust.maxhealth.tech/mcp
Skills: document-signing
Proxy Smart
Deployable SMART-on-FHIR proxy for a healthcare organization.
Endpoint https://api.proxy-smart.com/mcp
Skills: smart-app-onboarding, smart-launch-debugging
Chess
Real-time chess: a shared lobby and live games, playable from any MCP client.
Endpoint https://chess.maxhealth.tech/mcp
Skills: chess-coaching
Connecting a client
Every endpoint above speaks Streamable HTTP, so any client that supports remote MCP servers can use it by URL. Add the endpoint in your client's MCP settings; if it is a sign-in endpoint, the client opens a browser window once and remembers the grant.
6 of them also ship as installable plugins, bundling the connector with its skills. In a client that supports plugin marketplaces, add max-health once and install them by name — the plugin directory lists every entry with its install command. (It used to point at /products, which lists products rather than plugins and could only say “plugin available”.)
Common questions
What is the Model Context Protocol?
MCP is an open standard for connecting AI assistants to outside systems. Instead of every assistant needing a bespoke integration for every tool, a system exposes one MCP server and any MCP-capable client can use it. It standardises three things: tools the model can call, resources it can read, and prompts it can be given.
Why does MCP matter for health data?
Health records are scattered across portals, clinics and devices, and the interesting question is almost never "show me a table" but "what did that result say, and has it changed?". MCP lets an assistant read and cross-reference those records directly, under your account, instead of you copying values between tabs. Because the protocol is open, the same connection works in whichever client you prefer.
Is my health data used to train models?
No. An MCP server passes data to the client you connected, for the request you made. Max Health endpoints that touch anything account-scoped require you to sign in first, and access is limited to what that account can already see.
What is the difference between a connector, a skill and a plugin?
A connector is one MCP endpoint. A skill is a task-scoped instruction package that teaches a client how to do something well. A plugin bundles connectors and skills together, so installing one gives a client both the connection and the know-how.
What are MCP Apps?
MCP Apps is an extension for returning interactive interfaces rather than plain text or JSON. Endpoints that support it can answer with a real chart, table or viewer rendered inside the chat. Max Health builds these on its own open component system, Prefab.
Do I need a Max Health membership to use the MCP endpoints?
Not for the public ones, which need no account at all. Endpoints that reach your own data require sign-in, and what they can see follows whatever your account is already entitled to — a membership raises that level across products rather than unlocking the protocol itself.
Bring your records together first
An assistant is only as useful as what it can reach. A membership raises your level across every Max Health product.