Every clinician who evaluates Claude Cowork asks the same question: "Can it connect directly to our EHR?" The answer is yes — but the implementation path depends on your EHR vendor, your institution's IT infrastructure, and how much functionality you need on day one. This article covers all three integration tiers, from zero-configuration workflows that work immediately to full FHIR API integration with bidirectional data flow. While this guide focuses on clinical documentation broadly, the EHR integration setup is identical whether you are deploying for physicians or for nursing teams using Cowork for shift handovers and care planning.
This article is part of our complete Claude Cowork for Doctors guide. If you're setting up Cowork for clinical documentation workflows, start with the daily rounds workflow guide. For MCP architecture background, see the Claude Cowork Connectors Guide.
How Claude Cowork Connects to External Systems
Claude Cowork uses the Model Context Protocol (MCP) as its integration layer. MCP is an open standard developed by Anthropic that allows AI models to securely connect to external data sources and services. Think of an MCP server as a bridge: on one side is Cowork, on the other is your EHR system. The MCP server handles authentication, data formatting, and the API calls between them.
For EHR integration, an MCP server typically wraps your EHR's FHIR API and exposes patient data as structured resources that Cowork can read in real time — rather than requiring physicians to manually export and upload chart data. For the general MCP architecture explanation, see our MCP Protocol Guide.
The Three Integration Tiers
No EHR integration, no IT involvement, no waiting. Physicians export relevant chart sections from their EHR (as PDFs, text exports, or copied text), paste them into a Cowork canvas, generate documentation, and paste the outputs back. This is how most physicians start with Cowork — and it delivers 60–90 minutes of daily documentation savings before any integration is built.
The limitation is manual data loading. For high-volume floors, copying chart data for 20 patients before rounds adds 15–20 minutes to the pre-rounds workflow. Acceptable for individual physicians, less ideal for department-wide deployment.
Most EHR systems support scheduled data exports to a shared folder or SFTP location. Your IT team configures the EHR to export a physician's patient panel (daily notes, labs, medications) to a shared drive at a scheduled time — e.g., 5 AM before rounds. An MCP server for Cowork is configured to read from that folder, making patient data available in the canvas automatically without manual upload.
This delivers 80–90% of the benefit of full FHIR integration with significantly less complexity. The limitation is latency — data is only as current as the last export, not real-time. For daily documentation workflows, this is rarely a problem. For acute monitoring scenarios, full FHIR integration (Tier 3) is better.
Full integration connects Cowork to your EHR's FHIR R4 API via an MCP server, enabling real-time patient data reads and (for Epic) note write-back through the App Orchard API. The physician opens Cowork, selects a patient from their panel, and the canvas automatically populates with current demographics, diagnoses, medications, labs, and recent notes — pulled live from the EHR.
This is the highest-value configuration but requires the most infrastructure. For Epic, write-back (posting completed Cowork-generated notes directly to Epic without copy-paste) requires App Orchard access, which most hospitals already have. For Cerner and Meditech, write-back via API is also available but the specific configuration differs by version.
EHR-Specific Integration Details
Epic FHIR Integration
Epic's FHIR R4 API is the most mature healthcare FHIR implementation available. It supports read access to: patient demographics, problem lists, medication lists, lab results, imaging reports, clinical notes (including H&P, progress notes, and discharge summaries), and appointment data.
The MCP server for Epic integration uses OAuth2 with SMART on FHIR authorization, which means each physician authenticates once and the connection is maintained under their Epic credentials. No separate password management required.
Epic write-back requirements: Posting AI-generated notes to Epic requires your institution to be enrolled in the Epic App Orchard programme and to have the Note API scope enabled. Most large health systems already have this capability. Our deployment team confirms App Orchard access during the initial scoping call.
Cerner Millennium Integration
Cerner's SMART on FHIR implementation supports similar read/write patterns to Epic. The Cerner Millennium API exposes patient clinical data through standard FHIR R4 resources. Cerner's clinical documentation API supports note write-back for enrolled organisations.
Key difference from Epic: Cerner uses a different authentication flow and some FHIR resources are structured differently. The MCP server for Cerner must be built to Cerner's specific API documentation rather than a generic FHIR spec. Our MCP Server Development team handles Cerner-specific server builds as part of our deployment service.
Meditech Expanse Integration
Meditech Expanse (the current-generation Meditech platform) supports FHIR R4 with SMART on FHIR authentication. It provides read access to most clinical data resources. Meditech's write-back capability for AI-generated documentation is more limited than Epic or Cerner — the standard path for Meditech is read-only FHIR integration (Tier 3 for reads) combined with the clipboard workflow for note submission.
Older EHR Systems (Meditech 6.x, Allscripts, NextGen)
EHR systems without FHIR R4 support can use Tier 2 integration (export folder approach) as the primary connectivity method. Most legacy EHR systems support HL7 v2 message export or daily data dumps that an MCP server can read. This is less elegant than FHIR but delivers the same documentation workflow outcomes.
HIPAA Compliance Requirements for EHR Integration
Required before processing any PHI: Your organisation must execute a Business Associate Agreement (BAA) with Anthropic before using any real patient data in Claude Cowork. This applies to all tiers. Contact Anthropic directly or through your enterprise agreement to request the BAA. Our Cowork Security & Compliance guide covers the complete compliance framework.
Beyond the Anthropic BAA, EHR integration adds additional compliance requirements:
- Data transmission encryption: All data flowing between your EHR API and the Cowork MCP server must use TLS 1.2 or higher. All MCP servers built by our team implement this by default.
- Access logging: Patient data accessed through the MCP connector must be logged with physician identity, timestamp, and patient ID for audit trail purposes. This is typically handled at the EHR API level but must be verified during the integration review.
- Minimum necessary access: The MCP server's FHIR scopes should be restricted to the minimum data required. Don't request all FHIR resources — define the specific resources needed (observations, medications, clinical notes, etc.) and scope the OAuth2 credentials accordingly.
- Data residency: Confirm with Anthropic whether your Cowork deployment stores data in your required geographic region. This matters for organisations with EU data residency requirements or state-specific healthcare data laws.
Build vs Buy: Should You Build the MCP Server Yourself?
Clinical informatics teams with FHIR experience can build an Epic or Cerner MCP server in 2–4 weeks using Anthropic's open-source MCP SDK. The MCP specification is published at modelcontextprotocol.io and the server architecture is straightforward for any developer familiar with REST APIs and OAuth2.
For organisations without in-house clinical informatics development capacity, our Claude Cowork deployment service includes MCP server development, EHR API configuration, HIPAA compliance review, and physician training. A typical Epic integration project completes in 6–10 weeks from kickoff to production deployment.
The make vs buy decision usually comes down to: do you have developers familiar with both FHIR APIs and MCP who can dedicate 2–4 weeks to this project? If yes, build it. If not, a deployment partner is faster and lower risk.
Related Articles in the Series
Frequently Asked Questions
How long does a full Epic FHIR + Cowork integration take to build?
For organisations with App Orchard access already enabled and an internal clinical informatics developer, a read-only Epic FHIR MCP server can be built in 2–4 weeks. Adding note write-back adds another 2–3 weeks of testing and Epic's technical review process. Total project timeline for a production-ready Epic integration with our deployment team is typically 6–10 weeks including testing, HIPAA review, and physician onboarding.
Can Cowork connect to Epic MyChart for patient-generated data?
Epic's FHIR API includes patient-reported data accessible through MyChart — including questionnaire responses, symptom logs, and patient-entered medical history. This data is accessible under the same SMART on FHIR scopes as clinician-entered data, provided your Epic configuration exposes it. For care coordination use cases (e.g., reviewing patient-reported outcomes pre-visit), this data can be loaded into the Cowork canvas alongside clinical notes.
Does Claude Cowork store patient data from the EHR?
Cowork processes patient data within the context of your session. Under Anthropic's Enterprise terms with a BAA, Anthropic does not use your input data to train models. Data retention policies depend on your specific Anthropic Enterprise agreement — most hospital deployments configure zero data retention (conversation data is not stored after session end). Confirm your retention settings with Anthropic during the BAA process.
Our hospital is on Meditech 6.x (not Expanse) — what are our integration options?
Meditech 6.x does not support FHIR R4. Your integration options are: (1) Tier 2 — configure Meditech to export daily HL7 v2 files to a shared folder, and build an MCP server that reads and parses those files into the Cowork canvas; or (2) Tier 1 — use the copy-paste workflow with Meditech's PDF export function. Tier 2 is achievable but requires a developer familiar with HL7 v2 parsing. Our deployment team has built HL7 v2 MCP connectors for legacy EHR systems on previous projects.
EHR Integration Is an Architecture Problem. We Solve It.
Our Claude Certified Architects have configured Cowork + EHR integrations for Epic, Cerner, and Meditech environments. We scope the project, build the MCP server, conduct the HIPAA review, and get your physicians in production.