Skip to main content
AGO exposes an MCP (Model Context Protocol) server. Once connected, an AI assistant such as Claude, Claude Code or Cursor can read your conversations, knowledge base, tickets and analytics, and — if you allow it — update your knowledge and agents. You ask in plain language; the assistant calls AGO for you.

Connect your assistant

Your MCP address is:
Replace <your-tenant> with your subdomain (the part before .useago.com in your AGO URL).
  • Claude (web or desktop app): add AGO as a custom connector and sign in with your AGO account. See Connect Claude Web to AGO.
  • Claude Code, Cursor and other clients: create an API key in Developer > API Keys, then follow the client configuration in the MCP Server reference.
What the assistant can do depends on the permissions you grant (OAuth scopes or API key scopes). Start with read-only permissions; add write permissions only for the people who maintain knowledge or agents.

Start from a ready-made workflow

AGO ships ready-made workflows the assistant runs end to end. In Claude, open the + menu (or type /) and pick one from the AGO connector; in other clients, look for MCP prompts. You can also simply describe the job. Examples of plain-language requests:
Improve the Support agent: answers are too long and it should offer a ticket as soon as the user mentions billing.
Give me last week’s report for the Support agent.

Triage feedback into tasks

Use this recipe to turn a week of user feedback into a short list of tasks your team can act on, without duplicating work already on the board. Before you start: Tasks must be enabled for your workspace (contact AGO), and you connect as an active staff user with tasks:write and conversations:read. Add knowledge:read so the assistant can check existing knowledge update requests.
1

Start the workflow

In Claude, pick Triage feedback into tasks from the AGO connector. Optionally give a start date, an end date or an agent name. Without dates, it covers the last 7 days. You can also ask in plain language:
Triage last week’s feedback on the Support agent into tasks.
2

Let the assistant gather the evidence

The assistant reads the board’s statuses and the staff who can own tasks, lists the feedback for the period, and skips feedback that is already linked to a task. It groups the rest by root cause, reading the conversation when the feedback comment is not enough, and checks whether a knowledge update request already covers a gap.
3

Review the proposal

You get one proposed task per root cause: a title, what to do, a priority, and the feedback, messages, conversations or knowledge update requests it links. Nothing is created yet. Ask for changes: merge two tasks, drop one, raise a priority, assign an owner.
4

Approve

Once you approve, the assistant creates the tasks with an idempotency key each, so a retry never creates duplicates, and lists them with links to open them in AGO. New tasks appear on the board with the source MCP.
Run it weekly. Feedback already linked to a task is skipped on the next run, so you only see what is new.

Stay in control of changes

  • Read and write permissions are separate: an assistant without write permissions cannot change anything.
  • AGO tells the assistant to describe every change and wait for your confirmation before creating, updating or deleting. Deleting tools are flagged as destructive, so clients such as Claude ask you to approve each call.
  • Changes to an existing agent are prepared in its draft: end users keep the live agent until the draft is published. You can open the agent in AGO to review the draft, test it, undo a change or discard it.
  • Document and prompt edits are versioned: you can review and restore previous versions in AGO.

Troubleshooting

The assistant says it lacks a permission — the connection was not granted that scope. Reconnect (OAuth) or create an API key with the scope, then start a new conversation. A list comes back empty — nothing matched the filters for that period. Ask the assistant to widen the period or remove a filter. When a filter value does not exist (an agent name, a ticket status), AGO answers with the valid values instead of an empty list. AGO tools do not appear — check the MCP address and that your client supports the streamable HTTP transport, then restart the client.