Agent access in the dashboard
Agent access is a section of your workspace settings where you connect AI agents, see what each one may do, revoke connections and review what they did.
Connect an agent
- Open your workspace Settings and find Agent access.
- Copy the MCP server URL with Copy URL.
- Add it to your client, then sign in to TOSE when the client asks, pick the workspace and choose what to allow.
| Client | Steps |
|---|---|
| Claude | Settings, Connectors, Add custom connector, paste the URL. |
| ChatGPT | Settings, Apps & Connectors, Create (developer mode), paste the URL, then sign in with OAuth. |
| Cursor | Add {"mcpServers":{"tose":{"url":"<MCP server URL>"}}} to ~/.cursor/mcp.json. |
| VS Code | Run MCP: Add Server, choose HTTP and paste the URL. |
If the section says the MCP connector is not enabled, that server has it switched off. Agents can still use a workspace API key with the CLI or API.
What an agent can be allowed to do
The settings page shows the scope table: which scopes exist, which change production and who can grant them. The scopes are the same as in OAuth scopes. Members, billing, environment variables, databases and infrastructure are never available to agents, and write scopes need an owner or admin.
Your connected agents
The list shows the connections you authorized: client, workspace, scopes, last used, and expiry. Revoke signs that agent out at once.
Audit history
Your recent agent activity in this workspace lists every tool call an agent made with your connections, with time, tool, project, outcome and client. Arguments and results are not stored.
Agent tasks and Copy for agent
On a project page:
- Agent tasks lists deploys and rollbacks your agents asked for, with their status and links to approve or view the deployment. Nothing runs until you approve. See Tasks.
- Copy for agent copies safe project context to paste into an agent: slugs, links and the latest deployment. It contains no secrets.