Connected Agents (MCP)
Connected Agents let an AI assistant such as Claude, ChatGPT or Cursor work with your Cleotic brands through the Model Context Protocol (MCP). The assistant can read the same evidence you see in the app, and with your go-ahead it can triage opportunities, write Content Studio pages, add brands and prompts, and share reports.
A connection acts as you. It can do whatever your organisation role and brand access allow, and nothing more.
Connect
Every client uses the same endpoint, also shown in Settings → Connected Agents:
https://mcp.cleotic.ai
You sign in to Cleotic in the browser and choose the organisation the connection works in. Choose your client:
In Claude on the web or desktop:
- Open Customize → Connectors.
- Select + Add, then Add custom connector.
- Name it
Cleoticand enterhttps://mcp.cleotic.aias the remote MCP server URL. - Continue, then sign in to Cleotic and approve the connection.
On Team and Enterprise plans, an owner first adds the connector in Organization settings → Connectors. Members then select Connect beside it in Customize → Connectors. See Anthropic's custom connector guide.
After you approve the connection, the client discovers Cleotic's tools. To use another organisation, add a second connection and choose that organisation when you sign in.
What to ask
Ask in your own words; the assistant picks the tools. These are common jobs, with the tools they use and what they cost. Reading is free, and the assistant should confirm anything that costs before it goes ahead.
Get a health brief
How is Acme doing this month? What needs attention first?
Uses list_brands, get_brand_overview and list_opportunities. Figures come with their basis: the answers and models behind them, the likely range, and whether a change is real. Free.
Explain a change
Why did Acme's visibility drop in ChatGPT last month?
Uses get_brand_overview to see whether the change is real and which models moved, then list_answers filtered to that model, get_answer for the answers themselves and list_citations for the sources behind them. Free.
Triage the opportunity inbox
Go through Acme's open opportunities. Assign the content gaps to me and snooze the rest until Monday.
Uses list_opportunities and get_opportunity to read the evidence, then update_opportunities to move, assign, snooze or dismiss, and comment_on_opportunity to leave notes. Free.
Write a page for a content gap
Write a page for Acme's strongest content gap.
Uses create_page from the gap, get_page for the interview questions only you can answer, answer_page_questions to record your answers, then draft_page, and mark_page_published once the page is live. Your client may also offer this as the Plan a page prompt. Model budget for the questions and plan; draft_page uses one Content Studio generation, so the assistant should confirm before it drafts.
Fix and verify a site readiness issue
What's the most important readiness fix for Acme's site? Give me the change to make.
Uses get_readiness for the action plan and get_readiness_finding for the drafted fix. Once you've deployed it, act_on_opportunity records the fix as shipped and verify_readiness_fix re-checks the pages. A re-check from the inbox is free once every 10 minutes; a re-check of a finding uses one of the plan's monthly re-checks.
Add a brand
Start tracking northwind.example.
Uses start_brand_setup, which reads the website and proposes competitors, a market, a segment and prompts. The assistant then reviews the proposals with you through get_brand_setup, and approve_brand_setup creates the brand with the ones you keep. Model budget to read the site; the brand, segment and kept prompts use plan capacity, and their first answers use model budget.
Track more questions
Suggest ten questions UK buyers ask about expense software, and add the best five to Acme.
Uses suggest_prompts, ranked by AI search demand, then save_prompt for each one you keep. One model call for the suggestions; each new prompt uses an active prompt of plan capacity and gets its first answers straight away.
Share the results
Make a report of Acme's last quarter and give me the link.
Uses create_report, which returns the public link; revoke_report kills a link. One report of plan capacity.
See the tool reference for every tool.
Access and limits
| Sign-in | OAuth in the browser. Cleotic API keys and X-API-Key aren't accepted. |
| Organisation | One per connection, chosen at sign-in |
| Permissions | Your organisation role and brand access, checked on every call. Where you can only view a brand, the assistant can read it but not change it. |
| Terms | Until you accept Cleotic's current terms, only the documentation tools work |
| Rate | 600 calls an hour per organisation. Over it, calls fail with HTTP 429 and a Retry-After header. |
| Response size | Each response is at most 32 KB. Lists return shorter pages to fit; the assistant follows next_cursor for more. |
| Dates | The last 30 days unless the assistant asks for a window, of at most 180 days. Plans with limited history return the days they keep. |
| Plan | Writes respect your plan's capacity, Content Studio allowance and model budget, and are refused once a trial ends |
Cleotic doesn't list or revoke connections yet: to disconnect, remove Cleotic from the client. Content Studio writes never publish to your CMS, and Cleotic never sends outreach.
Troubleshooting
Check that the endpoint is https://mcp.cleotic.ai and that the client uses remote MCP over Streamable HTTP.
Related
Updated