Claude is good at the writing and reasoning work that piles up around a private practice: drafting letters of medical necessity and prior-auth appeals, summarizing a chart history before a visit, building a spreadsheet for billing or scheduling. None of that needs any setup. You can do it today in the Claude app with nothing connected to your EHR. Connect Claude to an MCP-capable EHR like Eureka and it goes further: it can read today's schedule and a patient's actual chart, then draft the letter or the refill order as a proposal you approve, not a task it just runs on its own.
What can I use Claude for today, without connecting anything?
The everyday case for a psychiatrist running a solo practice is turning a page of notes into a finished document. You paste in what you know, Claude writes a first draft in roughly the voice you'd use, you edit it and send it. No EHR connection, no API key, just the Claude app or claude.ai.
| Task | Example prompt |
|---|---|
| Letter of medical necessity | "Write a letter of medical necessity for continued twice-weekly psychotherapy for a patient with treatment-resistant depression who partially responded to two prior SSRI trials." |
| Prior-auth appeal | "Draft an appeal for a denied prior authorization on extended-release methylphenidate for adult ADHD, noting the patient's inadequate response to the immediate-release formulation first." |
| Patient-education handout | "Write a one-page, plain-language handout on what to expect during the first two weeks of starting sertraline." |
| Practice policy document | "Draft a 24-hour cancellation and no-show policy for a solo, cash-pay psychiatry practice." |
Each of these works from what you type or paste in. Claude has no memory of your patients between sessions and no view of your schedule unless you give it one, which is what the next section covers.
What changes when Claude can see your schedule and chart?
On its own, Claude only knows what you put in the prompt. Connected to your EHR over MCP (Model Context Protocol), it can read live data instead: today's appointments, a specific patient's chart and timeline, open orders. That's the difference between "draft a letter about a patient like this" and "draft the letter of medical necessity for my 2pm, using her actual diagnosis history and medication trials."
Setup for Claude specifically, Claude Code, Claude Desktop, and claude.ai connectors, is in connecting Claude to your EHR. Eureka's own MCP server sits at api.eureka.md/mcp: about 37 tools across roughly 20 domains, each split into a read tool and a separate write tool, so a doctor can allow only the read half and get an agent that can look but not touch. The full tool and resource list is in the MCP reference. Two prompts ship with the server out of the box: /prep-day for a morning brief before clinic starts, and /post-visit to walk a visit from note through finalize and handoff.
Even connected, the clinical boundary holds. Prescribing stays in the app, notes carry your signature, and when Eureka's built-in assistant suggests a reschedule or refill, it arrives as a one-click proposal in your practice's Slack channel for you to approve. Appointment reminders and recall outreach are the exception: once you configure them, those send on their own, because there's no clinical judgment in reminding a patient about a booking they already made.
Not every EHR exposes this kind of access yet. Most cash-pay and legacy systems still don't; our EHR MCP roundup has the current state across the field, and running a private practice with AI agents makes the broader case for connecting any AI agent, Claude or otherwise, to a practice.
Is this new?
Anthropic launched Claude for Healthcare in January 2026, aimed at consumer health questions and enterprise connectors rather than at a single doctor's own chart and schedule. Pointing Claude at your own patients through an EHR's MCP server, the way this page describes, is a separate setup, and it's the one the rest of this page is about.
Who decides, Claude or you?
What we tell our own customers is this: Claude drafts, you decide. It writes the letter, works through a differential, summarizes the labs. You read it, edit it, sign it. That holds whether Claude is unconnected and working off what you paste in, or connected to your EHR and working off live chart data. The drafting gets better with more context. The ownership boundary doesn't move.
Ask around in psychiatry and therapist communities online and you'll hear the same line drawn over and over: automating the admin work is welcome, automating the clinical judgment is not. That's the line Eureka builds to.
Is this HIPAA compliant?
Eureka signs a business associate agreement with every practice we work with, and that covers PHI moving through Eureka itself, including our MCP server. It does not automatically cover a general Claude subscription you connect to your own patient data on your own. When you connect Claude, ChatGPT, or any other AI tool to PHI yourself, your practice is the covered entity, and you're responsible for having the right agreement with that vendor. Most consumer AI subscriptions don't include a BAA. Anthropic and OpenAI both launched healthcare-specific offerings in January 2026 that may cover this, so check directly with your vendor before sending PHI through a tool you set up yourself. This isn't legal advice, just the shape of the risk. One practical mitigation: allow only the read-only half of an MCP server for anything that touches patient-authored text, so the worst case an outside AI tool can do is see data, not act on it.
To connect your own Claude to your practice's schedule and chart data, request an MCP API key at eureka.md/support and follow the steps in connecting Claude to your EHR.