What a connected agent can and cannot do

Twenty five tools, covering most of what you would otherwise do by clicking. Worth reading once before you hand an agent a token.

A connected agent can:

• call whoami to check who it is acting as
• run search, and list your work and your notifications
• list organisations, projects and the items inside a project
• read an issue, a feature or change request, or a task
• create issues, feature and change requests, client requests and tasks
• update those items, change a status, assign an item and add a comment
• delete a task
• create, list and update your personal tasks, and add notes to them

Assignment works in plain language. Asking it to assign something to a colleague by name resolves the name for you, so you never hand your agent a user ID and you never go and find one yourself. If the name is ambiguous it says so rather than guessing.

The permissions are the honest part. An agent holds exactly the permissions of the user whose token it carries: the same role, the same organization and project membership, nothing more. It cannot see a project you are not a member of, and it cannot see an internal item you could not see. Every write is attributed to that user, and every title or description edit is snapshotted into the same history as an edit made in the browser, so you read the change the same way either way.

Three limits worth knowing up front:

• the token’s role needs team workspace access, so a client role cannot use MCP at all
• file attachments are not supported over MCP, and still go through the web app
• a revoked token, a deleted user and a suspended account all fail as the same plain unauthorised response, on purpose, so nobody can probe for which one it was

A failed tool call comes back as a readable result rather than a crash, so an agent can tell you “you are not a member of that project” instead of stalling.