Function Health
Read a summary of your Function Health lab results and your nutrition protocol from an AI agent. Three read-only tools, three named read scopes, no write path. Function's own docs state individual lab values are never sent through the connector.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Connect Function Health via MCP
https://services.functionhealth.com/ai-chat/mcpWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Function Health Tools & Capabilities (3)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- The connector returns counts, not values. Function states individual lab results are never sent. An agent cannot tell you what your LDL actually was, only that it fell outside range. For numbers, open Function.
- We read no tool schemas or annotations from the server. The endpoint 403'd every anonymous request, so every parameter and annotation on this page comes from Function's documentation rather than a
tools/listresponse. - Neither published list gives the wire identifiers. Anthropic lists display labels, Function's docs give descriptive headings plus three call names in prose. The two sets share no string.
- The published surface has no history tool.
includeTrendsis documented as a parameter on the overall summary rather than a separate longitudinal tool, so trend depth is whatever that flag returns — the docs do not say how far back it reaches. - No clinician notes, imaging or diagnoses are in scope. MRI & CT Scans appears only as one of 24 biomarker *categories*. Nothing in the surface returns a report, an image, or a physician's interpretation.
- HIPAA coverage is conditional and Function says so. Its privacy policy states HIPAA does not apply to all personal information it processes. Which regime applies depends on whether you or an employer bought the membership.
- The free-text
contextparameter is undocumented in substance. Function describes it only as additional context for personalization and does not state retention or storage. - No rate limits are published. Function's documentation names none, and we could not observe any behind the edge block.
- The documentation is a single page with no version history beyond 1.0.0, dated January 2026. There is no changelog to watch for surface changes.
- The connector is not medical advice, and Function's own site states it does not provide medical advice, diagnosis or treatment. An agent reasoning over out-of-range counts is generating text, not practising medicine.
Frequently asked questions
Function's documentation says it does not. The docs state that individual lab test values or specific results, biomarker IDs, email addresses and detailed personal health information are never sent through the connector, and that it retrieves summary-level data only — counts of in-range and out-of-range biomarkers, category names, and nutrition goals. We could not independently verify this, because the endpoint is auth-gated.
No write path exists in the published surface. All three tools read, all three OAuth scopes begin with read, and Function's documentation states the connector has read-only access and cannot modify your information. It also publishes readOnlyHint true and destructiveHint false. We could not read those annotations off the wire ourselves, so they are Function's claim rather than our observation.
Not uniformly, and Function says so directly. Its privacy policy states Function may operate as a HIPAA business associate but that HIPAA does not apply to all personal information it processes. For a consumer membership bought directly, the data is generally consumer health data under state law rather than HIPAA-protected. Function publishes a separate Consumer Health Data Privacy Policy covering that path.
Three named read scopes, one per tool. The server's RFC 9728 descriptor advertised read:health_summary, read:biomarkers and read:action_plan on 2026-08-22. Each maps to one tool, so the consent screen genuinely separates the three surfaces. No scope grants writing anything, and the identity-only scopes on Function's Auth0 tenant are separate from these three.
A Function membership with completed lab results, and a paid Claude plan. Function's documentation lists both as prerequisites and notes that connectors require a Pro or Max subscription. If your lab results are still being processed the tools return nothing, which Function documents as a distinct troubleshooting case rather than an error condition.
Because an edge layer blocks the request before authentication is evaluated. On 2026-08-22 the endpoint returned a generic Google 403 HTML page to an anonymous handshake, with no www-authenticate header, and returned the same body to a plain GET. That is a WAF block rather than an auth challenge, so it says nothing about the server's auth posture.
One parameter deserves attention. The nutrition tool accepts a free-text context parameter for personalization, which Function's docs describe only as additional context. Everything else is a filter over data you already hold. Treat that field as the one place where text you supply, rather than summary counts, leaves the conversation.
Yes, from either side independently. Function's documentation states you can revoke authorization from your Function account settings and separately disconnect or remove the connector from Claude at any time. It also notes you may request data deletion under applicable privacy laws. Function recommends disconnecting when you are not actively using the connector.
Sources
- Function Health Connector for Claude documentation (retrieved 2026-08-22). The three tools, their parameters, the 24 categories, the scopes, the shared-data lists, troubleshooting and the annotation table all come from this page. · retrieved 2026-08-22
- Function Health privacy policy, HIPAA and business-associate wording (retrieved 2026-08-22) · retrieved 2026-08-22
- Function Health Consumer Health Data Privacy Policy (retrieved 2026-08-22) · retrieved 2026-08-22
- Function Health terms of service (retrieved 2026-08-22) · retrieved 2026-08-22
- Live RFC 9728 protected-resource metadata at
https://services.functionhealth.com/.well-known/oauth-protected-resource/ai-chat/mcp, returning the three read scopes and the Auth0 authorization server (2026-08-22) · retrieved 2026-08-22 - Live Auth0 authorization-server metadata at
https://auth.functionhealth.com/.well-known/oauth-authorization-server, showing PKCE support and identity-only tenant scopes (2026-08-22) · retrieved 2026-08-22 - Live endpoint probe: anonymous initialize and GET against the endpoint, both returning an identical Google edge 403 with no authentication challenge (2026-08-22) · retrieved 2026-08-22
- Function Health
robots.txt, carryingContent-Signal: search=yes,ai-train=no,use=reference(retrieved 2026-08-22). We cite and link rather than restating Function's prose, per that signal. · retrieved 2026-08-22 - Anthropic Connectors Directory entry — , read from our directory snapshot dated 2026-08-16 · retrieved 2026-08-16
- Function support — <mailto:support@functionhealth.com> · also published as <mailto:hello@functionhealth.com> in the connector documentation
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Healthcare
- Developer
- Function Health
- Tools
- 3
- Domain
- services.functionhealth.com
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.