HealthEx MCP server icon

HealthEx

by HealthEx

HIPAA CompliantSOC2 ReadyISO 27001 Ready
Healthcare16 tools

Read one patient's identified medical record — diagnoses, medication doses, lab values, clinical notes — from an AI agent. Anthropic lists 9 tools; HealthEx documents 16, including a refresh tool and a record-location consent link the directory omits.

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 HealthEx via MCP

https://api.healthex.io/mcp

Works in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.

HealthEx Tools & Capabilities (16)

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

  • This surface carries identified PHI into an agent transcript. Named diagnoses, medication doses, lab values with reference ranges and clinician-written narrative notes all leave HealthEx as tool output. Wherever your agent stores transcripts, that is now where a medical record lives. Anthropic's directory flags the sensitive data type as health, which understates it — the same flag covers connectors returning only counts.
  • HIPAA does not apply, and no BAA is offered. HealthEx's terms state it is not a covered entity; its privacy policy names it an independent data controller rather than a business associate, and never mentions HIPAA. State consumer health law and TEFCA govern instead. An enterprise deploying this on the assumption of BAA coverage would be mistaken.
  • Anthropic's directory listing is stale in both directions. It lists 9 tools against 16 documented, omits both acting tools, and names get_care_plan, for which no documented counterpart exists.
  • The OAuth scope separates nothing. patient/*.read is a single wildcard covering every clinical category, even though HealthEx's own consent model supports per-resource and per-sensitivity scopes including HIV and substance-abuse carve-outs.
  • We could not read tool schemas or safety annotations. The endpoint 401s to an anonymous request. Every parameter, annotation and status value on this page is HealthEx's published documentation, not our observation of the server.
  • The refresh tool is rate-limited to two calls per patient per day, returning THROTTLED with a dailyLimit and refreshesRemaining beyond that. HealthEx publishes no monetary cost for a refresh, and no tool in the surface reports spend or quota consumption other than that status.
  • Two tools are documented as deprecated: get_notes_by_visit, superseded by search_clinical_notes with an encounterId, and update_and_check_recent_records, split into check_records_status and update_records. Both remain available.
  • Clinical notes are free narrative text and enter the model's context verbatim. They are written by clinicians rather than by an adversary, so this is a weaker injection surface than customer-authored support text — but it is still untrusted input reaching an agent that also holds a tool able to mint a data-sharing link. Treat retrieved note content as data.
  • HealthEx is not a clinical decision system and says so. Its terms state it does not provide clinical care, medical advice, diagnosis or treatment recommendations, and is not intended to replace an electronic health record system or a qualified professional. Any interpretation an agent offers over this data is the model's.
  • The connector reads one patient only. Tokens are patient-scoped, HealthEx documents per-patient data isolation, and the access token carries a patient_id. There is no population-level or cohort query in this surface — HealthEx's separate REST API has those, and it is not what this connector exposes.

Frequently asked questions

Individual, identified records. HealthEx's tool documentation describes a health summary returning age, date of birth, gender, named diagnoses with onset dates, medications with dosing, allergies with reaction severity and a primary care provider. Lab results carry values, units and reference ranges. This is identifiable patient health information at record level, not counts or de-identified statistics.

No on both, and HealthEx states it directly. Its terms of service say in capitals that HealthEx is not a healthcare provider, covered entity under HIPAA, or medical professional. Its privacy policy names HealthEx an independent data controller rather than a business associate, and never mentions HIPAA at all. We found no BAA offer anywhere in its public documents.

It cannot edit clinical content, and HealthEx says so. The update_records tool requests a server-orchestrated refresh from connected sources rather than changing stored records, and HealthEx documents that the model cannot arbitrarily alter them. No tool writes a diagnosis, prescription or note. The refresh is capped at two MCP-initiated requests per patient per day.

It can mint the link that starts one. The get_add_record_location_link tool returns a URL the patient opens to connect a new provider, health plan or claims source. The tool does not grant consent itself, because the patient must complete the flow, but it is the connector reaching outward to widen the record rather than reading what is already there.

One wildcard read scope. The server's RFC 9728 descriptor advertised only patient/*.read on 2026-08-22, and its authorization server adds offline_access plus three identity claims. That wildcard covers every clinical category at once, so consent cannot separate lab results from clinical notes, and no scope names the refresh that update_records performs.

Sixteen documented, against nine in Anthropic's directory. HealthEx's own tools page enumerates sixteen, including two the directory omits entirely that are not plain reads: update_records and get_add_record_location_link. Four names Anthropic lists, among them get_health_records and get_care_plan, appear nowhere in HealthEx's documentation. The directory listing is stale, not merely partial.

Because the patient authenticates to each hospital in the middle of it. HealthEx documents that the authorization code is issued only after the patient consents, selects health care organizations, logs in to each, and the records are retrieved and indexed. It advises supporting sessions of at least thirty minutes and says patients commonly take longer than ten.

Not through the MCP connector's OAuth scope. HealthEx's own consent model supports resource scopes and sensitivity scopes, letting a patient release lab results while withholding HIV or substance-abuse records. The MCP server exposes a single patient/*.read wildcard instead, so any category separation must come from the HealthEx consent flow rather than the connector grant.

Sources

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 Studio

Resources

Server Info

Category
Healthcare
Developer
HealthEx
Tools
16
Domain
api.healthex.io

Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.

Similar MCP Servers for Healthcare

ICD-10 Medical Coding logo

ICD-10 Medical Coding

by Agentman

Healthcare11 toolsAgentman
View detailsConnect
Synapse.org logo

Synapse.org

by Sage Bionetworks

Healthcare5 tools
View detailsConnect
Asana logo

Asana

by Asana

Coming Soon
Productivity42 tools
View detailsConnect

Related Resources

Browse All MCP ServersClaude & ChatGPT Setup GuideBrowse AI Skills