Connect Rankfor data to an internal AI assistant

What must the client support?

Confirm that your organization permits external MCP connections and that the selected client supports authenticated HTTP servers. Product names alone do not establish connector support in every account or deployment. Follow the client’s current setup documentation and your organization’s policy for which project data may be sent to the assistant. Choose one non-sensitive, authorized project for the initial read test.

Rankfor’s MCP server endpoint is https://mcp.rankfor.ai/. Configure that endpoint as an authenticated HTTP MCP connection, using the client’s supported secret or authorization-header field for a bearer token. Exact connector menus vary by client. A web browser visiting the server root is not an MCP connection test; use the client’s discovery and tool-list flow to establish that the protocol and credentials work.

Where do credentials come from?

In the Rankfor app, open your profile at /user/me and find Access tokens. Under Add a new access token, optionally give the token a useful name and choose Create access token. The full secret is shown once. Store it in the approved secret location for that client and keep it out of shared chats, documents and screenshots.

Each colleague should use their own authorized setup.

The connector’s customer tools read existing records. The personal token authenticates on your behalf. Share the credential only with approved clients and review the access granted to the integration. Revoke a token through the profile when the integration is retired or the credential needs replacement, and update the authorized client through the normal credential process.

What is a meaningful first test?

Ask the assistant to list the projects it can access, then identify the intended project explicitly. Read available engines and existing scan status before requesting an analysis. A useful prompt is: “For this authorized project, summarize the latest existing Index record and include its engine and measurement date.

If the record is missing, say so.” The answer should describe returned data without pretending a new scan ran.

Next, inspect a stored Answer Trail or content plan. Customer tools include list_answer_trails, get_answer_trail_details, list_content_plans, get_content_plan and get_content_asset. Draft text is requested explicitly with draft="ai" or draft="human" for the relevant existing asset. The assistant should distinguish a stored draft from its own suggested rewrite and retain the record’s scope in the explanation.

How do you accept the integration?

Compare one assistant summary against the same record in the app. Check project identity, date, selected engine, missing fields and any important qualification. Try a request for unavailable data and confirm the assistant acknowledges the gap. Do not test success by asking it to generate a new Rankfor brief: reading an existing plan and creating a saved asset are different capabilities.

Record connection owner, authorized use, acceptance result and credential review date without storing the secret in the checklist. Use the public Rankfor playbooks to choose the business workflow around the data. The completed integration should reduce the effort of reading existing evidence while leaving consequential product changes visible in the app.