Agent Access

Agent Access lets your own AI agent (such as Claude Code, Cursor, or another MCP client) read your Privas AI workspace, so it can answer your own questions about your setup. Privas AI acts as the server here; your agent connects to it.

Where to configure it

Open your domain, then go to Settings > Channels & Integrations > Agent Access.

Create a key

1. Create a new API key.
2. Copy it when it is shown.

The key is shown once, at creation. Store it somewhere safe, because it cannot be shown again. If you lose it or it is exposed, revoke it and create a new one. Revoking takes effect immediately.

Connect your agent

Point your MCP client at the Privas AI endpoint:

https://privas.ai/api/mcp

Authenticate with your key. Your agent can then read your workspace configuration.

Read-only by default

By default, keys are read-only: your agent can read your settings, not change them. This is the safe default for connecting an external tool.

If writing is enabled

Where write access is available and enabled, a change is never applied blindly. The agent first requests a preview of the exact change, and the write only proceeds against that preview, so it cannot skip the review step. Every change is recorded in a change log you can see on this page, and a change can be undone.

Next step

For operational control of the widget itself, see Widget Kill Switch.