REST API, Webhooks, and MCP Server

API keys

Programmatic access is an Enterprise plan feature. Create and revoke keys from the API keys panel, available from the sidebar inside the tool and from the account dashboard. Keys are prefixed rgsk_ and are shown once at creation, so store them in a secrets manager. Every request authenticates with a Bearer header: Authorization: Bearer rgsk_...

REST endpoints

The version 1 REST API exposes four endpoints over your cloud-saved reviews:

GET /api/screen/v1/projects lists your projects. GET /api/screen/v1/projects/{id} returns one project in detail, including its records and decisions. POST /api/screen/v1/projects/{id}/records appends a batch of records to a project, de-duplicating like an in-app import. POST /api/screen/v1/projects/{id}/decisions writes screening decisions in bulk, keyed by record id.

Write limits: both POST endpoints accept up to 5,000 items per request, and a project holds at most 50,000 records through the API; oversized batches are rejected with an explanatory error rather than partially applied.

1. Create a key in the API keys panel and store it securely.
2. List your projects: curl https://researchgold.org/api/screen/v1/projects -H "Authorization: Bearer rgsk_..."
3. Fetch one project by id from the list response.
4. Push new records or write decisions with the POST endpoints, batching at up to 5,000 items per call.

Webhooks

Instead of polling, register an HTTPS endpoint and the tool will push events to it. Three event types exist: records.appended (new records were added to a project), decisions.updated (screening decisions changed), and project.updated (project metadata changed). You can register up to 10 webhooks and subscribe each to any subset of events.

Every delivery is signed. The request carries an X-RG-Signature header containing an HMAC-SHA256 signature of the payload, computed with the secret issued when the webhook was created. Verify the signature on your server before trusting the payload, and toggle or delete hooks from the same management panel.

MCP server for AI assistants

MCP (Model Context Protocol) is an open standard that lets AI assistants call external tools directly. The screening tool speaks it at POST /api/screen/mcp, authenticated with the same Bearer API key, so an assistant like Claude can work with your reviews conversationally.

Five tools are exposed: list_projects, get_project, search_records, append_records, and set_decisions. That is enough for an assistant to find the right review, look up records, add search results, and record screening decisions you dictate, while your account's plan limits and audit trail still apply to everything it does.