The product

Deploy an agent

When your blueprintis complete, one click is enough to push the agent to Anthropic Managed Agents. Here's what happens under the hood, what it costs, and how to react when it breaks.

The deployment flow

When you click Deploy, Forgent does this:

  1. Serializes your blueprint into JSON conforming to the Managed Agents API.
  2. Sends the request to /api/ma-proxy, our internal proxy that attaches the platform credentials only for the duration of the call.
  3. The proxy forwards to the Anthropic Managed Agents endpoint with the required beta header.
  4. Anthropic provisions the agent on its infra and returns an agent_id.
  5. Forgent stores the association { user, agent_id, blueprint } in the database and redirects you to My Agents.
Why a proxy rather than browser callsThe Managed Agents API (in beta at the time of writing) doesn't send the right CORS headers, so direct calls from the browser fail. The proxy solves that, and keeps the platform credentials out of the client JS.

The possible statuses

  • Provisioning — Anthropic is spinning up your agent (a few seconds to 1 minute).
  • Ready — the agent is deployed and ready to be run.
  • Error — something failed on Anthropic's side. The details are shown in the error modal.
Nothing to set upBuilding, deploying, and running an agent are all free — funded by the platform. There is nothing to plug in and no usage bill (see Getting Started).

How much it costs

Nothing. Forgent funds building, deployment, and every run:

  • Building the blueprint (the Builder's ~5 questions): free — the Builder assistant is funded by the platform.
  • Deployment itself: free, it's just an API request.
  • Running the agent: free — every run is funded by the platform on Claude Fable 5.

Common errors

"Tool not available"

You selected an MCP server (Notion, Gmail...) that your Anthropic account hasn't connected yet. Go to the Anthropic console, authorize the MCP in question, then rerun the deployment.