Resources

Before your AI agent creates a task or answers a question about your project, it usually reads something first: who it's connected as, what it's allowed to do, the actual task or page it's about to work on. That's what a resource is. You never fetch these yourself; your AI client (Claude, Cursor, whatever you're using) pulls them in automatically while it works.

What your agent can see

• Its own identity and permissions. Before trying anything, a well-behaved agent checks what it's actually allowed to do, so it doesn't waste a step attempting something the connection was never granted.
• Your workspace. Which workspace it's connected to; every connection belongs to exactly one.
• A specific task, project, wiki page, sprint, or team, once the agent has an ID for it, for example after you mention a task number or paste a link.
• Workspace settings and notifications, mostly relevant for an agent managing the workspace itself rather than doing day-to-day task work.
• Admin-only context (the member directory, other connected agents and their status) if it's connected as a human with workspace-management rights. A task-scoped agent never sees this.

Reading is still permission-checked

Reading any of this goes through the same permission check as an actual write would. If the connection doesn't have rights to see a task, it can't read that task's details either; resources aren't a way around the same access rules your MCP tools already follow. See MCP Tools for how that permission model works.

Agent Skills: packaged guides, not project data

Alongside your project's actual data, TAM also exposes ready-made guides for common workflows (sprint planning, triage, and similar) as their own kind of resource. These are the same for every connection; reading one doesn't need any workspace access, since there's no project data in them. See MCP Prompts for how an agent puts one of these guides to work.