Connect Vara to AI Assistants

Connection URL

Every client uses the same URL. Nothing to install and no API key to create: sign in to Vara when prompted and click Allow.

https://varaapp.site/api/mcp

Before you start, make sure API access is enabled for your account (a trial or a subscription, the same switch as API keys). You can check and enable it in Settings → API. Without it, authorization completes but the connection is refused with a message pointing to Settings → API.

Claude (claude.ai on the web and desktop)

1. Open Settings → Connectors → Add custom connector.
2. Enter any name (for example Vara) and paste the connection URL above.
3. Click Connect, sign in to Vara on the page that opens, review the access and click Allow.
4. Back in Claude, try asking "List my Vara experiments".

Claude Code

Run this in your terminal:

claude mcp add --transport http vara https://varaapp.site/api/mcp

Then type /mcp in Claude Code and select vara to authorize. Your browser opens the Vara sign-in and authorization page; click Allow and return to the terminal.

Cursor

On a computer with Cursor installed, click the button below to add Vara:

If the button does nothing, add this to mcp.json yourself, save, then sign in to Vara when prompted and click Allow:

{
  "mcpServers": {
    "vara": { "url": "https://varaapp.site/api/mcp" }
  }
}

Other clients

Any client that supports remote MCP with OAuth sign-in can connect with the same URL. For the local server or API key setup, see the MCP reference.

Access: read-only or read and write

The authorization page shows the app asking for access and the address it returns to after you approve. You choose the access level:

• Read-only: view your experiments, sub-experiments, run logs, plans, references and materials.
• Read and write: everything in read-only, plus create and edit these records, upload attachments, and delete plan items, attachments and reference links.
• With either level, core data (experiments, sub-experiments and run logs) cannot be deleted.
• A connection cannot read your AI provider keys, account settings or password.
• The access token the client receives can also call the Vara /api/v1 REST API directly, with the same access gate, scope and rate limits as an API key. The LLM gateway (/v1/chat/completions, /v1/models) does not accept it.

If the app name or return address on the authorization page is not the one you just started connecting, click Deny.

How to disconnect

• In Settings → API, find the app under Connected apps and click Disconnect. This takes effect immediately: the app can no longer access your Vara data and must be authorized again to reconnect.
• You can also remove the Vara connector in your AI client. Removing it only in the client does not cancel the authorization on the Vara side right away, so disconnect in Settings as well if you want access to stop immediately.

Where your data goes

• When you ask the AI assistant to do something, it calls Vara tools, and the Vara data those tools return (for example experiment names, steps, parameters and run notes) is sent to that AI platform and becomes part of your conversation with it.
• Whether that platform uses the data for training, and how long it keeps it, depends on your settings with that platform and its privacy policy. Vara does not control this.
• Vara only sends data to the client you authorized, not to any other app. Content already in a conversation has to be deleted on that AI platform.

See the Privacy Policy for details.

Attachment limits

A remote connection cannot read files on your computer. When the AI assistant uploads an attachment, it can pass a public file link (up to 25MB) or upload a small file directly (up to 3MB). Drag larger local files into the Vara web app instead.

Common problems