Run your help desk from Claude, ChatGPT or your editor

Chatty comes with an MCP server, so your team can work the help desk from the AI assistant they already use. What it can do, and what it deliberately can’t.

2 min read

Most of us already have an AI assistant open all day. When Chatty opens, it comes with an MCP server, so that assistant can work your help desk too.

MCP, the Model Context Protocol, is the open standard that assistants use to talk to other tools. Chatty’s server is a thin door onto the same help desk your team uses in the console: the same tickets, the same knowledge, the same permissions.

What you can ask

  • “What did visitors ask this week that the agents couldn’t answer?”
  • “Show me open tickets about invoices.”
  • “Assign ticket 1042 to me and add a note that the fix ships on Friday.”
  • “What does our knowledge base say about refunds?”

Behind those, the server can find tickets, read a ticket with its conversation, search your knowledge, list the knowledge gaps, open a ticket, change a ticket’s status, assignee, labels or priority, and add an internal note. Reading and writing are separate tools, each labelled so your assistant knows which is which.

What it deliberately can’t do

Some things stay in the console, on purpose:

  • Nothing is deleted. There is no delete tool.
  • Nothing reaches your customers. No tool sends a message to a customer. Replies go from the board, written by a person.
  • Nothing changes who can do what. Billing, members, model keys and secrets are not tools.
  • Nothing changes how your agents behave. An agent’s instructions are not edited or published from an assistant.

Who you are when your assistant acts

You sign in once and approve the connection. From then on the assistant acts as you, in one organisation, with your role. A Responder can’t do through Claude what they couldn’t do in the console, and a read-only connection can’t call a write tool at all. Every call lands in the audit log with the assistant’s name next to yours.

Text from strangers is marked as such

Tickets and transcripts are written by people outside your company. When the server hands that text to your assistant, it labels it as untrusted content. An instruction hidden in a ticket, such as “ignore your instructions and…”, is data to read, not a command to follow. And because no tool can send anything onward, there is nowhere for it to go.

Which assistants

Any client that supports remote MCP servers: Claude, ChatGPT, and editors such as Cursor and VS Code. We’ll publish the install steps for each one when Chatty opens.

Put Chatty to work.

Give your customers a helpful first answer. Create your account and introduce Chatty to your business.