Skip to main content
Once you’ve created a Deployment, either your own or one an admin set up for your project, you can attach it to any AI Model step in an agent so the model can call its tools while generating a response. This is separate from adding an app to a Gateway. See Deployments vs. Gateways if you’re not sure which one you need.

​
Attaching a Deployment

  1. Open your agent and select the AI Model step you want to give tool access to.
  2. In the step’s side panel, open the MCP Servers tab.
  3. Click Select MCP server deployments. A searchable list shows every Deployment available to you: anything created in your current project, plus any Deployment that isn’t tied to a specific project.
  4. Check as many Deployments as you want to use on this step. There’s no limit on how many you can attach.
  5. Save the agent.
Only Deployments you have access to appear in the list. If one you expect to see is missing, confirm it is to your current project or to all projects.

​
Creating a Deployment Without Leaving the Canvas

If you have admin permissions, a Create new deployment option appears below the selector so you can pick an app and set it up on the spot, without navigating away from your agent. Anyone else building an agent can only choose from Deployments that already exist. See Gateway/Deployment Creation for the full setup flow, or ask an admin to create one for your project if you don’t see the option yourself.

​
Choosing Which Tools This Step Can Use

Click into an attached Deployment to see every tool it has, each with its own on/off toggle, plus Enable All and Disable All for quick changes. A running count shows how many tools are active out of the total available. This selection is scoped to the step you’re editing, not to the Deployment itself. Turning a tool on or off here doesn’t change what’s enabled on the original Deployment, and different steps, in this agent or any other, can each enable a different subset of the same Deployment’s tools. That means you can reuse one Deployment across several agents while giving each one only the access it actually needs.
As with Gateways, every tool you enable adds to what the model has to read before it can respond. Turn on only the tools a given step actually uses, even if the underlying Deployment has more available.

​
Authentication

If a Deployment’s credentials are Personal, its detail view shows an End-user credentials banner, and every tool toggle stays disabled until you click Authenticate and connect your own account, even if someone else already connected the same app. Deployments with Tenant level credentials show a Tenant credentials banner instead and work for everyone the moment the Deployment is attached. See Tenant vs. Personal Level App Credentials for how these levels work.
If an agent with a Personal-credential Deployment is shared with other users, each of them needs to authenticate their own account before that Deployment’s tools work for them. Sharing the agent doesn’t share your sign-in.

​
Removing a Deployment

Open the MCP Servers tab and remove the Deployment’s card. This only detaches it from this step. It doesn’t touch the Deployment itself, so it’s still there to attach to this or any other agent later.
If a Deployment attached to a step gets deleted elsewhere, the step shows a warning in its place instead of failing silently. Remove the stale reference and attach a replacement if the agent still needs those tools.

​
Testing Your Tools

There’s no separate preview for which tools are reachable. Airia relays the tools listed by the origin server, and some servers, even after being authenticated, list but restrict access to certain tools depending on your account permissions. Instead, run the agent in the Playground and send a message that should trigger one of your enabled tools. When the model calls it, a tool execution card appears in the response showing which tool ran and whether it succeeded, so you can confirm your setup before publishing.