# Use Salisan from AI apps (MCP) (/docs/en/guides/ai-apps)

Claude, ChatGPT, Claude Code, Cursor and other apps that connect to remote MCP servers can work in Salisan for you. A connected app signs in as you and has exactly your access: it sees the patients, records, appointments and tasks you can see, and changes only what your role allows. You manage your apps in **Settings › AI apps**, which everyone who may use the Agent has.

## Turn on AI apps for the practice [#turn-on]

AI apps are off until an owner or admin turns them on; it needs the **Manage AI credits** permission. Everyone else sees a note that owners and admins decide.

1. Open **Settings › AI apps**.
2. Under **Practice**, switch on **AI apps for** your practice. They are on right away.

If you start connecting an app while AI apps are off, the screen the app opens offers **Turn on AI apps** as well.

Turning AI apps off asks first, then disconnects every app for everyone in the practice. Members can connect again once it is back on.

## Connect an app [#connect]

1. Open **Settings › AI apps**. Under **Which app do you use?**, choose your app, or **Other apps**.
2. Under **How to connect**, choose **Copy** beside the **Server address**. For Claude Code, copy the **Command** instead, and for Cursor the **Configuration**. The page shows the address for your practice's region.
3. Follow the steps shown for your app. In Claude, for example, you open **Customize › Connectors**, choose **Add**, then **Add custom connector**, and paste the address.
4. Sign in to Salisan when the app asks. If your practice requires two-factor or single sign-on, sign in that way.
5. On the consent screen, read what the app may do. Optional: tick **Only let … read**, so the app can look things up but not change anything.
6. Choose **Allow**. You return to the app, and it appears under **Your connected apps**.

<Callout type="warn">
What you open in an AI app is sent to that app's provider. Before you choose **Allow**, check that your practice has the right agreement with that provider. The consent screen reminds you of this.
</Callout>

The consent screen shows the logo of a known app, such as Claude, ChatGPT, Claude Code or Cursor, and where you return after signing in. If it says the app registered itself, Salisan cannot confirm who runs it. Choose **Cancel** unless you just started connecting it yourself.

## What an app can and cannot do [#limits]

- **Your access, nothing more.** The app uses your role's permissions. It never sees patients who keep their record away from AI, including patients under 18 unless the practice allows it; see [Data and compliance](/docs/en/guides/data-and-compliance#ai-minors).
- **The app asks before changes.** Salisan keeps no approval list for AI apps: each app asks you before a change by its own settings. Under **Your connected apps**, each app says where those settings are, such as **Customize › Connectors › Salisan** in Claude.
- **Changes land as drafts where Salisan uses drafts.** Record changes wait until a clinician keeps them. Letters stay drafts until sent, care plans until activated and templates until published.
- **Some settings stay in Salisan.** Apps cannot change billing, sign-in rules, members and roles, the AI app switches, exports, retention or publishing.
- **Practice switches.** Owners and admins can switch on **AI apps can only read**: every app can look things up and answer, but cannot draft, book, send or change anything, whatever members allowed. **AI apps can message patients** decides whether apps may send emails, forms and home reading requests. It is hidden while apps can only read.
- **Every action is recorded.** The audit log names the AI app and that it acted for you, so its actions are told apart from Salisan's own Agent. See [Access and audit](/docs/en/trust/access-and-audit).

An app never makes a clinical decision. Check what it drafts before you sign or send it, as you would with the [Agent](/docs/en/guides/agent). For what reaches each provider, see [How the AI works](/docs/en/trust/how-the-ai-works#ai-apps).

## Disconnect an app [#disconnect]

1. Open **Settings › AI apps**.
2. Under **Your connected apps**, choose **Disconnect** beside the app.
3. Confirm with **Disconnect**. The app loses access right away. To use it again, connect it again from the app.

Each connected app shows **Can make changes** or **Read only**, and when it was last used. Owners and admins also see **Everyone's connected apps** and can disconnect any member's app the same way. Removing a member from the practice also ends their apps' access; see [People and access](/docs/en/guides/people-and-access#remove).

A connection is a sign-in, so it also ends on its own:

- when the practice's **Sign in again after** time has passed since you chose **Allow**, 24 hours by default; see [Practice settings](/docs/en/guides/practice-settings#sign-in-policy);
- when you change or reset your password, choose **Revoke other sessions**, remove a passkey or turn off two-factor.

Then connect again from the app. Salisan asks you to sign in and shows the consent screen again.

Your Agent preferences apply in connected apps too; see [Agent settings](/docs/en/guides/agent-settings#memory).