Connect Rivly to Claude, ChatGPT or Cursor

Step 1: create a key

1. In Rivly open “MCP & agents” on the left. The key belongs to the project selected at the top left.
2. Choose the rights: “read only” is enough for analyses and text drafts. “Read and write” additionally lets the assistant tick off steps of a recommendation and extend the measurement plan.
3. Click “Create key”. The key starts with rv_live_ and is shown in full exactly once.
4. Copy the key straight away and store it in your password manager. Later you only see the shortened form. If it is lost, revoke it and create a new one.

Only the account owner can create and revoke keys. Every project has its own key: for a second brand, switch the project at the top left and create another one there.

Step 2: enter the credentials

Choose your assistant. In every case you replace <YOUR-KEY> with the key you copied, then you restart the assistant.

Claude Desktop

Open the settings, section Developer, and edit the file claude_desktop_config.json. Insert the entry, quit Claude Desktop and start it again.

{
  "mcpServers": {
    "rivly": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.rivly.de", "--header", "Authorization: Bearer <YOUR-KEY>"]
    }
  }
}

Cursor

Create or extend the file mcp.json in the folder .cursor of your user directory, insert the entry and restart Cursor.

{
  "mcpServers": {
    "rivly": {
      "url": "https://mcp.rivly.de",
      "headers": { "Authorization": "Bearer <YOUR-KEY>" }
    }
  }
}

ChatGPT

1. In ChatGPT open Settings → Connectors and add “Rivly”.
2. Server URL: https://mcp.rivly.de
3. Authentication: Bearer, followed by your key.

Your own agent

Every agent that speaks the Model Context Protocol reaches Rivly with a POST request at the same address. The call below lists the tools and thereby checks whether the key works.

curl -X POST https://mcp.rivly.de \
  -H "Authorization: Bearer <YOUR-KEY>" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

What the agent can do

Twenty-five tools are available, eighteen for reading and seven for writing. The writing ones work only with the right “read and write”, otherwise Rivly answers that the key may only read.

1. get_overview (read): Rivly Index, channel values and the “In short” of the current week.
2. get_ai_answers (read): AI answers word for word per buying question and system, with sources.
3. get_rankings (read): Google positions, Google AI Overview and search volume per term.
4. get_competitors (read): competitors with values from all four channels.
5. get_recommendations (read): open recommendations with steps, points and status.
6. get_core (read): profile, offer, search queries with volume, buying questions and competitors with the reason they were added, from the core of the project.
7. get_map (read): market map of the plan: fields with status, demand, leader and own page, plus the measures with their id.
8. get_site_audit (read): website check: technology, sitemap, llms.txt, access for AI providers and page potential.
9. get_page_brief (read): page brief for a measure or a field: heading, sections, customer questions, schema and llms.txt line.
10. get_llms_txt (read): draft of the llms.txt with the pages of the core, plus the lines for pages that are not online yet.
11. get_structure (read): structure of the website against demand: demand without a page, pages not on place 1 and topics without demand.
12. get_live_check_history (read): History of the AI Live Check: questions with the answers of ChatGPT and Gemini, source app or MCP, rest of the weekly quota.
13. get_ads (read): Ads channel: active ads of the brand and its competitors from Meta and Google, with themes and winners.
14. get_press (read): Press channel: editorial articles of the last 30 days, share of coverage and media that name only competitors.
15. get_backlinks (read): Backlinks in the Website Check: linking domains against the competitors, strong links, gaps and missing directories, collected monthly.
16. get_business_profile (read): Google Business in the Website Check: own listings checked against the website, place in Google Maps and reviews.
17. get_history (read): History per calendar week: Rivly Index and the channels GEO, SEO, ads and press, plus events such as a new measurement basis.
18. get_measure_plan (read): Measurement plan: which buying questions and search queries Rivly measures every week, with the last result and free slots in the tariff.
19. update_recommendation (write): tick off a step, schedule it or report it as done.
20. add_prompt (write): add a buying question or search query to the measurement plan.
21. check_page (write): check one page of the website right away: reachable, sitemap, linking, search query, schema and llms.txt.
22. mark_action_done (write): report an active measure as done, the next measurement checks the effect.
23. ask_live_check (write): Ask ChatGPT and Gemini one question live, with the same weekly quota as the chat in the app.
24. promote_live_check (write): Take a question from the Live Check into the measurement plan as a buying question so it is measured every week.
25. create_page_brief (write): Have a new page brief written for a measure or a field, with the same monthly quota as the button in the Fahrplan.

The methodology as a resource

Besides the tools, Rivly provides a description at rivly://methodik of how to read the Rivly Index, the four channels and the measurement weeks. An assistant that supports resources can read it before the first analysis and then does not misread the numbers.

Example: building a page

This is how an assistant with the right “read and write” turns a measure from the Fahrplan into a finished page. You say, for example: “Take the first open measure from the Fahrplan, fetch the page brief and build the page. Then check it and report the measure as done.”

1. get_map: the assistant reads the Fahrplan and picks a measure. If none is active yet, it sets one to “in progress” with update_recommendation.
2. get_page_brief: it fetches the page brief with heading, sections, customer questions, schema and llms.txt line. If there is none yet, create_page_brief has it written. After about a minute it appears in get_page_brief.
3. The assistant builds the page. With access to the code of your website, for example in Cursor, it creates the page right there.
4. check_page: after publishing it checks the page right away: reachable, sitemap, linking, search query, schema and llms.txt.
5. mark_action_done: it reports the measure as done. The next measurement checks the effect.

Security

The key is valid only for this project and only for the rights you chose. Other projects of your account stay out of reach. An assistant is never allowed to do more than you: whatever your role in the account may not do, Rivly also refuses through the key.

Every request is logged, with time, tool and assistant. Rivly does not store the wording of the requests and answers. To prevent misuse, the IP address of a request stays in the log for 90 days, after that Rivly deletes it.

“Revoke” disconnects all connected assistants immediately. For further access you create a new key and enter it again.

If it does not work

1. Error 401: the key is revoked or entered incompletely. Check that the word Bearer stands in front of it with a space and that the key was copied in full, otherwise create a new one.
2. Error 429: more than 60 requests in one minute per key. The assistant can repeat the request after a minute.
3. “This key may only read”: the tool writes, the key only has read rights. Switch it to “read and write” under MCP & agents.
4. “Quota reached”: ask_live_check counts like the chat in the app per week, create_page_brief like the button in the Fahrplan per month. Rivly tells the assistant the date from which there is room again. A brief that is already written stays free to open with get_page_brief.
5. The assistant does not find Rivly: usually the restart is missing. Quit the assistant completely and open it again.