Browser, API and MCP

Choose a client for the same research contract, and use the canonical setup documentation.

The browser, public API and MCP expose the Research engine through different interfaces. Keep the exact query and its result receipt when moving between them.

Browser: explore and review

Use the Research Workbench to formulate a question, inspect the proposed conditions, review counts and outcome distributions, and open available replay evidence. The workspace also owns saved studies and live alerts, subject to account access.

API: integrate with software

Use the API reference for authentication, grammar, examples, limits and response contracts. Read the registry before submitting field IDs. Receiving a proposed query from interpretation is separate from executing it.

MCP: research with an agent

Use the MCP setup guide for current clients, connection methods and tools. Connect an AI agent manages the account connection. This workflow guide deliberately does not duplicate tool lists or configuration snippets that can drift from the setup reference.

The agent should show the exact proposed question and scope before execution. Reading an existing report and creating a new computation are different operations with different access and allowance rules; consult the current API documentation.

Keep one source for each detail

Use the Reading library for measurements, API docs for machine contracts, and these guides for the workflow. The Developers page helps choose an integration path.