> ## Documentation Index
> Fetch the complete documentation index at: https://docs.asknow.no/llms.txt
> Use this file to discover all available pages before exploring further.

# Connect an AI assistant

> Connect Claude, ChatGPT or Cursor to your Ask workspace over MCP: setup, access levels and what each tool needs.

Ask has an MCP server. Once you connect your own AI assistant to it, you can ask things like "Which topics did customers ask about most last week?" or "Run the tests for our shop agent", and the assistant reads or changes your workspace for you.

<Note>
  The server URL is `https://app.nordic-software.com/mcp`. You sign in with your normal Ask account; there is no API key to copy.
</Note>

## 1. Turn on access for the workspace

Access is off by default. An admin of the workspace chooses what a connected assistant may read under **Settings → Connected assistants**. The setting applies to every member's assistant.

| Level | What an assistant can read |
| - | - |
| **Off** | The workspace's name, id and slug, that access is off, and your role. Nothing else. |
| **Agents and reports** | Agents and their numbers, conversation metadata (when, channel, agent, outcome, rating), topic counts and open cases. No message text, titles or escalation notes. |
| **Including conversation text** | Everything above, plus transcripts, case titles and escalation notes. |

<Warning>
  What the assistant reads is sent to the assistant's provider, under that provider's terms. At **Including conversation text** we remove email addresses and phone numbers we can recognise before sending, but this is best effort, and names are not removed. Start with the lowest level you need.
</Warning>

## 2. Add the server to your assistant

<Tabs>
  <Tab title="Claude">
    1. In Claude, open **Settings → Connectors** and choose **Add custom connector**.
    2. Name it `Ask` and paste `https://app.nordic-software.com/mcp` as the URL.
    3. Choose **Connect** and sign in to Ask in the window that opens.
  </Tab>

  <Tab title="Claude Code">
    ```bash theme={null}
    claude mcp add --transport http ask https://app.nordic-software.com/mcp
    ```

    Then run `/mcp` in Claude Code and choose **ask** to sign in.
  </Tab>

  <Tab title="Other clients">
    Any client that supports remote MCP servers over HTTP with OAuth works. Add `https://app.nordic-software.com/mcp` as the server URL. The client finds the sign-in page by itself from the server's OAuth metadata.
  </Tab>
</Tabs>

## 3. Try it

Try asking your assistant "Which Ask workspaces can you see?". It calls `list_workspaces` and lists each workspace with your role and its access level. Every other tool takes a `workspace` argument (its id or slug), so if you belong to more than one, say which one you mean.

## Roles and writes

Your role in the workspace still applies when you work through an assistant.

* **Any member** can use the read tools that the access level allows: agents, conversations, topics, cases, templates, memories and procedure exports.
* **Admins only** can use the knowledge, tests, feedback and recommended-action tools, and every tool that changes something: procedures, templates, memories, knowledge documents and tests.

Writes are careful by design:

* An imported procedure is created switched off, and an update never switches one on. You switch it on yourself.
* Templates, memories and knowledge documents take effect at once, since they have no draft. Replacing or deleting a knowledge document returns its old text, so you can put it back.
* `run_tests` starts billed test runs. The assistant is told to ask you before calling it.

Every call, including one refused for access, is written to the workspace's audit log with the assistant it came through. Free-text arguments such as a search query are left out of the log. Calls are rate limited per person and workspace.

## What your assistant can do

Once connected, your assistant shows each tool it can use with a description, so you don't need to learn them. They fall into these groups:

* **Reports**: agents and their numbers, conversation search and topics, open cases.
* **Conversations**: full transcripts, when the workspace allows **Including conversation text**.
* **Agent setup**: procedures, templates, memories and knowledge documents, which admins can also change.
* **Quality**: agent tests and their runs, feedback and suggested improvements, for admins.

Admin tools also need the workspace's access to be at least **Agents and reports**.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.