Airia Deployed Agents
Turn agents built in Airia Studio into MCP tools any Airia agent or external client can call.
Airia Deployed Agents is different from the rest of the catalogue: instead of connecting to an outside app, it connects to your own tenant. It turns agents you've built in Airia Studio into individual MCP tools that any MCP client can call directly, including another Airia agent, Claude Desktop, or Cursor.
It's a Featured catalogue server, pinned near the top of the server picker when you build a Gateway or Deployment. Like any catalogue server, an admin can still restrict or remove it for your organization through Server Management.
Prerequisites
- At least one agent with an Active version published.
- That agent's Tool & MCP interface turned on. See Tool & MCP Interface for how to name the tool, write its description, and optionally define input and output schemas.
An agent without the Tool & MCP interface enabled won't show up as a tool through this server. You don't need to touch the Gateway or Deployment side to add or remove an agent: it's picked up automatically as soon as you enable or disable the interface on the agent itself.
Connecting Airia Deployed Agents
Add Airia Deployed Agents to a Gateway or Deployment the same way as any other server. See Gateway/Deployment Creation.
| Method | How it works | Setup |
|---|---|---|
| Passthrough (recommended) | Airia forwards your own sign-in, so tool calls run as you with your existing permissions. | Nothing to configure. |
| API key | Every tool call runs as whichever identity the key belongs to — you, or a service account with its own roles. | Generate a key from API Keys in your account settings and paste it in during setup. |
Whichever method you use, the agents that show up as tools depend on the access level behind that connection. Two people connecting through the same Gateway with passthrough auth may see different tools if their permissions differ.
What Each Tool Includes
Once connected, any agent exposed through the Tool & MCP interface becomes a callable tool with:
- A name and description you defined, so callers know what it does and when to use it.
- An input schema, either the default single-input format or a custom one you defined, so callers know exactly what to send.
- Automatic version management: calling the tool always runs the agent's latest published version, so republishing an agent updates the tool immediately without touching your Gateway or Deployment configuration.
Two Ways to Call It
- From an external client. Point Claude Desktop, Cursor, or any other MCP client at a Gateway that includes this server, and your deployed agents show up as tools it can call.
- From inside Airia. Add it to a Deployment and use one agent as a tool inside another, without leaving the platform.
Conversation Memory and MCP Sessions
If an agent you're calling this way uses conversation-scoped memory, for example to remember a chat ID or thread across turns, that continuity is tied to the MCP connection itself rather than anything you pass in. As long as the calling client keeps the same session open, repeated tool calls share the same conversation memory.
If the calling client disconnects and reconnects, that counts as a new session, so conversation-scoped memory starts fresh. This is expected: a new chat session in Airia itself wouldn't carry over memory from a previous one either.
If you need memory that survives reconnects, use User-specific or Global scoping instead of Conversation-scoped. See Memory for the full breakdown of scoping options.