Skip to main content

Attaching Capabilities

The Capabilities tab is where an agent gets a model to run on and everything it can call or draw on while running — organized into four sections on its left-hand sub-navigation.

Before you begin​

Attaching a connection, tool, MCP source, skill, prompt, or guardrail requires that item to already exist in its registry, or you to have permission to create one on the spot. See Managing the Capability Registries.

Connect a model​

  1. Open the Capabilities tab, then select Model & tools.
  2. Under AI resource & model, select a connection from the Connection field. If none exist yet, select Manage connections to create one.
  3. Studio shows which model the selected connection uses. Model and sampling parameters are configured on the connection itself — edit them from the connection's advanced options, not here.

Attach tools​

  1. Still on Model & tools, under Tools, select one or more tools in the field that attaches tools from the registry.
  2. For each attached tool, set its run mode: Auto, Require approval, or Let agent decide.
  3. Expand a tool to preview its actions and required parameters.

Attach MCP sources and A2A registries​

  1. Select the MCP & A2A registry sub-tab.
  2. Use the field that links MCP sources from the registry to attach one or more.
  3. If an attached MCP source needs its own OAuth connection, a MCP connections section appears. Select Connect, complete sign-in in the new tab, then select Refresh here.
  4. Use the field that links A2A registries from the registry to attach agents this one can delegate to.

Attach skills and a prompt​

  1. Select the Skills & prompts sub-tab.
  2. Use the field that links skills from the registry to attach one or more skills.
  3. Optionally turn on Also autodiscover every organization skill to let the agent discover and load any other skill visible to its organization, in addition to the ones you attached.
  4. Use the field that links a prompt from the registry to attach a single prompt.

Attach guardrails​

  1. Select the Guardrails sub-tab.
  2. Use the field that links guardrails from the registry to attach one or more.

Guardrails run at the stage they were configured for: an Input-stage check can block a run before any model call; Tool use and Output checks are recorded but never block the run.

How you know it worked​

Each selection appears as a tag in its field, and the top bar's badge changes to reflect an unsaved draft. Select Save draft to keep it.

Next steps​