Using the ClaudIA MCP

When to use

• You want to upload, organize and review ClaudIA's content by chatting with your AI assistant (Claude, Cursor or another MCP client)
• You want to test answers and understand why ClaudIA made a decision without opening the dashboard
• You are in onboarding and want to speed up the initial setup

Prerequisites

• The Cloud Humans MCP connected to your assistant. The step-by-step guide is in How to connect the Cloud Humans MCP Server to Claude
• Your Cloud Humans user. The assistant has exactly the same permissions you have on the platform: whatever you can't do in the dashboard, it can't do either

About this article

To see the MCP details on the platform, click "Configurações do Perfil" (Profile Settings) > MCP / API. The full list of tools is in the API Documentation.

Below are examples of requests you can make to your assistant. Use them as a starting point and adapt them to your operation. For what to do in Cloud Chat, see How to use the Cloud Chat MCP.

How it works

Upload and organize content

• "Import these FAQs into ClaudIA's knowledge base": creates up to 200 contents at once. If any of them fail, the assistant shows which ones and why, without losing the ones that succeeded
• "Import the content from the site ajuda.empresa.com": registers the site as a content source and runs the import. It happens in the background and may take a few minutes
• "Add the tag Payments to all contents about bank slips"
• "Disable the content for the promotion that has already ended": the disabled content stays saved, but is no longer used in answers
• "Which tags and topics are in use in the knowledge base?"

Test and improve answers

• "Ask ClaudIA how to cancel an order and show me the answer": runs the real ClaudIA, with your prompts, contents and transfer rules, and shows what the customer would receive and where the ticket would go. No message reaches the customer and the conversation appears in the Playground inbox
• "List the suggestions from the "Centro de Melhorias" (Improvement Center) and approve the ones that make sense": these are suggestions of new content for questions the knowledge base didn't answer, with how many conversations fell into each gap
• "Which contents does ClaudIA suggest rewriting and why?": suggested rewrites for low-performing contents, along with the diagnosis
• "Change the initial greeting to: Hi! How can I help?"

Understand what happened in a conversation

• "Why did ClaudIA transfer ticket 1234?": accepts the conversation ID in Cloud Chat or the ticket ID in your helpdesk. Shows the closing reason and the reasoning behind the decision
• "Which contents did ClaudIA use to answer the last message of conversation 1234?"
• "Who changed the project settings this week, and what was there before?": every change is recorded with the previous value, so it can be undone

Specialist Agents

• "Register our order system's MCP for the agents to use"
• "List the tools available to the agents"
• "Create a draft refund agent that uses the order lookup tool"
• "Chat with the refund agent draft to test it"

Notes

• The tests ("Ask ClaudIA..." and "Chat with the draft...") run the real model and consume LLM tokens billed to your account
• If your user has access to more than one project, say in the conversation which one to use
• To test from the dashboard, see How to test ClaudIA in the Playground