With the Herodesk MCP server, a manager can set up an AI agent from their AI assistant instead of clicking through the AI agent screens. You need a connected assistant with write access, see What is the Herodesk MCP server?, and the manager or admin role. Agents can see the list of AI agents, but not their configuration.
What your assistant can do
| Task | Example prompt |
|---|---|
| See an agent's full setup | "Show me how our AI agent Sam is configured." |
| Create an agent | "Create an AI agent called Sam for our webshop, learn from our website shop.dk and connect our Shopify integration." |
| Rename or describe it | "Rename the agent Sam to Nova and change its industry to SaaS." |
| Change the personality | "Make Nova more formal, use emojis never and keep answers standard length." |
| Draft a personality from your team | "Analyse how our team writes in the Support inbox over the last 90 days and suggest a communication guideline for Nova." |
| Decide where it answers | "Let Nova draft replies (assist) on our live chat and email channels." |
| Handover and exclusions | "Never let Nova answer emails from @supplier.com, and hand over without replying when a customer asks for a human, except in live chat." |
| Delete an agent | "Delete the agent Test bot." |
Creating an agent
A new agent starts with:
- a friendly, casual personality with short answers, which you can change afterwards,
- the starter topics for its industry (e-commerce if you don't say otherwise),
- no channels: it does not answer anyone until you deploy it.
You can give it knowledge straight away: a website domain that is verified for your workspace, a Google product feed URL, and your webshop integrations. If one of them can't be used, for example a domain that isn't verified yet, the agent is not created and your assistant tells you why.
Personality and communication guideline
Your assistant can change the tone of voice, answer length, formality, emoji use, roles, the live chat welcome message and the communication guidance.
It can also draft a communication guideline from your team's own replies, like Generate from your conversations on the Personality page. At least 30 replies must match the filters you choose. Personal data is removed before the replies are analysed, and nothing changes on the agent until you ask the assistant to apply the draft.
Deploying: where the agent answers
For each channel the agent can be Off, Assist (drafts replies for your team), Review (a teammate approves each reply) or Auto.
Important: on a channel set to Auto, the agent replies to customers immediately, without a human checking. Your assistant asks before it chooses Auto, but read what it is about to change.
Only one AI agent can answer on a channel. If another agent already uses the channel, the change is refused and your assistant tells you which agent it is.
Your assistant can also set the handover replies, the inactivity check-in in live chat and WhatsApp, the senders the agent never answers, and the mode for conversations that match no topic.
Deleting an agent
Deleting an agent is permanent. Its knowledge (datasets, files and crawled pages) and its signatures are deleted too. Your assistant should always ask you before it deletes an agent.
Good to know
- Changes are recorded in the audit log under your name, the same as in the app.
- Knowledge and skills: Manage AI agent knowledge and skills with MCP
- Topics and instructions: Manage AI agent topics and instructions with MCP
- Testing before you go live: Test your AI agent with MCP
- Goals and fine-tuning: Track AI agent goals and fine-tune replies with MCP
- Copying an agent: Duplicate an AI agent with MCP