Agent access

For org owners and admins

Settings → Integrations → Agent access controls AI agents in your org. It has two separate switches, the list of tokens, and their activity. It’s in the desktop and web apps.

Agents in the app#

Agents in the app lets members connect Claude Code sessions to their own agents and talk to them in direct messages. Each session works as its owner, only in the projects they approve. It’s on to begin with.

Turning it off asks you to confirm. Every agent session in the org then stops getting messages and can’t work until you turn it back on. Messages sent meanwhile wait up to 7 days.

What members do with agents is in Working with agents.

Allow agent access#

Allow agent access lets members (not guests) create personal agent tokens, so AI tools such as Claude Desktop, Cursor or Claude Code can work in projects, pages and channels as them. Agents never see direct messages. It’s off to begin with.

Turning it off asks you to confirm. Every token in the org stops working at once. Tokens aren’t revoked: they work again when you turn it back on.

Under the switch, the page shows when it was last changed, and by whom.

Tokens in this organization#

Tokens in this organization lists every member’s tokens, newest first: who owns each, and its permissions and limits. Revoke stops a token for good.

Members manage their own tokens in Settings → Agent tokens. See Use other AI tools with an agent token.

Activity#

Activity lists what every token did, newest first. Filter it by Token to see one token’s calls, or choose All tokens.

For how agents are kept in check, see Safety and limits.