Openevidence MCP

What openevidence-mcp MCP server does

The openevidence-mcp MCP server connects MCP clients to OpenEvidence through a browser session belonging to the user. It runs locally over stdio and uses the saved profile created during setup rather than an OpenEvidence API token.

The server supports five main operations:

• Check whether the saved browser session is authenticated.
• List previous OpenEvidence articles, with optional pagination and search.
• Retrieve an article by ID.
• Submit a new research question, either blocking until completion or returning before the article is ready.
• Wait for an existing article to finish processing.

Article and history responses use normalized fields by default. Raw payloads can be requested explicitly for supported operations.

How it works

Run the server from the local project and configure an MCP client to launch it as a stdio process. Before making calls, authenticate through the supplied session-login command. That flow opens Chrome or Edge with a dedicated local profile; after signing in to OpenEvidence, the server reuses the same profile during its process lifetime.

For a short research request, oe_ask can wait for the result. For longer work, call it with wait_for_completion=false, keep the returned article ID, and pass that ID to oe_article_wait. Use oe_article_get when an article ID is already available. The repository also includes the openevidence_research_workflow prompt for clients that expose MCP prompts.

Setup and configuration

The openevidence-mcp MCP server requires Node.js 20 or newer, npm 10 or newer, an OpenEvidence account, and a Chromium installation supplied through Playwright. It supports macOS, Windows, and Linux.

A typical setup is:

1. Clone the repository and install its dependencies using the platform-specific setup script or the project instructions.
2. Run npm run login:session.
3. Sign in with your own OpenEvidence account in the opened browser.
4. Close the browser window and confirm the login in the terminal.
5. Run npm run smoke to check authentication and basic connectivity.

The default browser-profile location is ~/.openevidence-mcp/browser-profile on macOS and Linux, and %USERPROFILE%\\.openevidence-mcp\\browser-profile on Windows. The project also documents legacy login commands for development and browser-based troubleshooting.

Tools and capabilities

• oe_auth_status reports whether the saved local session is logged in.
• oe_history_list searches or paginates prior articles. Its default output reduces exposed personal data; raw output requires include_raw=true.
• oe_article_get returns normalized status, completion state, question, and answer text for an article. Raw data is optional.
• oe_article_wait polls an existing article until completion.
• oe_ask creates a research question and can optionally wait for its article to complete.

Submitting a question creates a question or article in the connected OpenEvidence account. Use original_article_id only when a follow-up should preserve an existing thread; omit it for a new question.

Limitations and notes

This is an unofficial project and is not affiliated with or endorsed by OpenEvidence. It does not bypass authentication, paywalls, or other access controls, and it does not provide an official API integration. The browser session remains local except for requests made to OpenEvidence, and the project states that it does not collect credentials or export cookies.

OpenEvidence access can depend on region, account eligibility, and OpenEvidence policy. The repository notes that availability may be centered on verified U.S. healthcare-professional access and that EU or U.K. access may be unavailable. Use the returned material as evidence-research context, not as medical advice, diagnosis, or a clinical order. Patient-specific decisions require appropriate human review and governance.