Eureka is a full EHR for solo private-practice doctors that ships a first-party MCP server, so tools like Claude Code, ChatGPT, and Codex CLI can read and act on real patient data instead of bolting onto the side of an app. The server runs live at api.eureka.md/mcp and authenticates with a per-doctor API key. Clinical judgment stays with the doctor: notes carry the doctor's signature and prescribing happens only in the app. The 14 questions below cover what practices, developers, and clinicians ask about it most.
Quick facts
| Item | Detail |
|---|---|
| Product | Eureka, an EHR and practice platform at eureka.md |
| MCP endpoint | https://api.eureka.md/mcp |
| Transport | Streamable HTTP, stateless JSON, no session state |
| Auth | Per-doctor API key: Authorization: Bearer eur_... |
| Tool surface | About 37 tools across roughly 20 domains |
| Clinical control | Doctor signs every note; prescribing stays in the app; the built-in assistant's suggestions arrive as one-click proposals |
| Works with | Claude Code, Claude Desktop, ChatGPT, Codex CLI, Cursor, any streamable-HTTP MCP client |
Is there an EHR that connects to Claude Code?
Yes. Eureka runs a first-party MCP server at https://api.eureka.md/mcp that Claude Code connects to directly with claude mcp add, using HTTP transport and an Authorization header carrying a per-doctor API key. Once connected, Claude Code can read patient charts, appointments, and notes, draft messages and visit notes, and schedule or reschedule visits with conservative defaults. See connecting Claude Code for the exact setup steps and the full tool list.
Is there an EHR that works with ChatGPT or Codex?
Yes. Eureka's MCP server speaks standard streamable HTTP, so it works with any client that supports the protocol, including ChatGPT's developer-mode custom connectors and OpenAI's Codex CLI through an [mcp_servers] block in its config file. The same API key and tool surface apply regardless of client. See connecting ChatGPT and the MCP server reference for setup details and current client syntax.
Which EHRs have MCP servers in 2026?
A handful, at different stages. Eureka runs a production, clinician-facing MCP server today. athenahealth announced a patient-facing MCP server in March 2026, still in pilot. CharmHealth announced an enterprise-oriented MCP server in February 2026. Healthie shipped "Dev Assist," an MCP server over its API docs for developers. Medplum, open-source, has official MCP documentation and a community server. See the full EHR MCP comparison for Epic, Elation, and Canvas Medical too.
Does SimplePractice have an API I can build on?
No. SimplePractice has no public API for individual practitioners, no webhooks, and no MCP server. Its AI features are a scribe add-on and a beta called "Care Aide," neither of which gives you programmatic access to your own data. If you want an agent that can read and act on patient records directly, you need an EHR built with an API and MCP server, the way Eureka is. See how EHR MCP servers compare.
Can an AI agent schedule my patients?
Yes. Through Eureka's MCP server, an agent working with the doctor's own API key can check open slots, book a new appointment, reschedule, cancel, and manage recurring visits. The defaults are conservative: a new booking leaves the visit unpaid rather than touching the patient's card, and the patient is not messaged unless explicitly requested. Eureka's built-in Slack assistant takes the more cautious route for its own suggestions: reschedules and cancellations arrive as one-click proposals the doctor approves. See AI agents for private practice.
Can an AI agent write my chart notes?
Yes, with a human signature required. Eureka's visit notes support AI drafting and voice dictation, so a doctor can dictate a visit and get a draft note back. The doctor still reviews and signs every note before it becomes part of the chart. No note is finalized by an agent on its own. See what Eureka's AI actually does for doctors for the full picture.
Can I automate appointment reminders and follow-ups?
Yes, and this is one of the few things Eureka runs fully autonomously once set up. Patients get automated reminders 48 hours before and the day of an appointment, over SMS and email. Recall and follow-up outreach works the same way: patients due for a check-in get contacted without a doctor triggering each message individually. Nothing clinical is decided in the process. See AI agents for private practice.
What can an AI agent actually do in Eureka?
An agent connected to Eureka's MCP server has roughly 37 tools across about 20 domains: patients, appointments, availability, billing, notes, orders, handoff, forms, messages, recall, reports, and more. Each domain splits into a read tool and a separate write ("_action") tool, so a harness can register read-only tools and get a genuinely read-only agent. See the full MCP server reference for the complete domain list and resource URIs.
How do I get an API key for Eureka's MCP server?
Eureka provisions the key for you. Ask in your practice's Slack channel or contact eureka.md/support, and Eureka issues a per-doctor key scoped to your own practice's data. Keys are sent as Authorization: Bearer eur_..., stored hashed, revocable at any time, and tracked by last use. There is no self-service key generation yet. See connecting Claude Code for what to do with the key once you have it.
Is it HIPAA-compliant to connect Claude or ChatGPT to my EHR?
Eureka signs a BAA with every practice. Connecting your own AI tool shifts responsibility to you: your practice becomes the covered entity for data sent to Claude or ChatGPT, and needs its own agreement with that vendor. Consumer subscriptions usually skip a BAA; enterprise and API tiers often include one, and Anthropic and OpenAI both launched healthcare offerings in January 2026. Registering only read-only tools cuts risk while that gets sorted. Confirm BAA status before sending PHI. See connecting Claude. Not legal advice.
Can the AI send prescriptions?
No. Prescribing is excluded from the MCP order tools entirely: e-Rx, including controlled substances through Eureka's RXNT integration, happens only in the app, under the doctor's own login. What an agent can do is surface the request: Eureka's built-in assistant turns patient refill requests into one-click proposals in Slack, and the doctor approves or declines each one. See what Eureka's AI does for doctors for the full boundary between drafting and executing.
What happens if the agent tries something Eureka does not support?
It gets a clear, structured answer instead of a silent failure or a fabricated result. Every tool validates its inputs and says exactly what is missing or out of bounds, and capabilities that are deliberately excluded, like creating a prescription through the order tools, are stated in the tool's own description so the agent knows to route the doctor to the app. See the MCP server reference for the full tool surface.
Do I need to know how to code?
No, for most setups. Connecting Claude Desktop, claude.ai, or ChatGPT uses a custom connector configured in settings rather than a command line, though the exact steps depend on the client and can change. Claude Code and Codex CLI are command-line tools, so someone comfortable with a terminal, often a technical contact at the practice, should handle that setup. Eureka provides the API key either way. See connecting Claude and connecting ChatGPT for current steps.
What does "AI-native EHR" mean?
It means the EHR itself is built to be operated by an agent. Eureka's data model, tool surface, and approval flow are designed together, so an MCP client can read a chart and act on it directly. Most "AI agent" products for healthcare are overlays that write back into a legacy EHR. See what AI-native means for a practice for the fuller argument.
Have a question this page didn't answer? Reach us at eureka.md/support.