openkp.org

What It Is

OpenKP is a local MCP server for Kaiser Permanente Northern California patients. It does not offer a hosted service and is not a standalone large language model. Instead, it lets MCP clients such as Claude Desktop, Claude Code, Cursor, and Zed read, compare, and—after confirmation—perform a small number of actions through the patient’s own kp.org session. Its positioning is closer to an “AI tool layer for patient-managed medical records,” turning the scattered results, notes, messages, plans, and orders in a traditional patient portal into an interface users can ask follow-up questions about.

Core Features and Use Cases

The project provides 27 MCP tools: 22 read tools, 2 write tools, and 3 housekeeping tools. The read scope covers profiles, insurance, PCP, care teams, implanted devices, access logs, test and procedure orders, preparation instructions, messages and attachments, lab results, imaging, PDFs, medications, allergies, appointments, past visits, visit notes, and after-visit summaries. Typical questions include: which lab values have been drifting over the past 18 months, how many online vs. in-person appointments there were last year, when third-party apps accessed which categories of medical records, which tests doctors have not yet completed, and how cardiology and primary care describe the same issue in their notes. On the write side, it only supports previewing and requesting mail-order refills, and previewing and sending non-urgent care-team messages; both require explicit confirmation.

Pricing, Privacy, and Integration

The text does not disclose commercial pricing or payment methods. The project is source-available on GitHub and uses the PolyForm NC 1.0.0 non-commercial license. Its privacy design is a major highlight: there is no OpenKP server, shared database, or remote credential store; credentials are kept in the system keychain or local environment, and submission flows have local audit logs. That said, the AI assistant must be able to see the parsed medical-record content in order to reason over it, so users still need to assess the risk of data leaving their environment when using cloud models such as Claude. Integration is based on the MCP standard, with Claude Desktop and Claude Code currently tested as clients. Installation involves a Python virtual environment, MCP configuration, and a one-time Chromium login; Windows also requires additional dependency steps.

Pros, Cons, and Best-Fit Users

Its strengths are a local-first approach, tools that closely match real patient needs, preview-and-confirm safeguards for write actions, and the ability to answer cross-record questions that ordinary portals do not handle well. The downsides are also clear: it has only been tested with KP NorCal, so geographic applicability is extremely narrow; it is a research tool rather than a clinical product; automated access may conflict with Kaiser’s terms; changes to portal endpoints can break functionality; and installation is not friendly for non-technical users. It is best suited to patients or research-oriented users with some command-line ability who want to review their own KP NorCal medical records. It should not be used as a substitute for medical decision-making.

Access from China

The text does not provide information about access from mainland China, network connectivity, or payment. Even if openkp.org or GitHub is accessible, actual use still depends on having a kp.org account, Kaiser NorCal membership, and access to MCP clients such as Claude. Chinese users typically will not have an applicable KP medical-record scenario, and may instead consider local medical-record organization tools or general-purpose knowledge-base solutions that support MCP.