Deployment

WorkflowsDeployment

Deploying publishes a workflow and gives it an address, so something other than the editor can run it. Nothing inside the workflow changes — the same blocks, the same Start block — but now an external request can reach it and start a run.

What you will learn

Deploy gives it an address

Publishing makes a workflow reachable from outside the editor at a stable, versioned address. The workflow itself does not change.

Calls flow into Start

An external request arrives at the same Start block you tested with, runs the chain, and the response goes back to the caller.

Pick a surface

Expose the one workflow as an API endpoint, a hosted chat page, or an MCP tool other AI systems can call.

A working workflow

Before deploying, you have a workflow that runs when you run it, in the editor. Deploying is the step that lets anything else run it too.

Here's the shape of the workflow you'll deploy — a Start block, a chain, and a response back to the caller:

Deploy gives it an address

Deploying publishes the workflow and gives it an address. Nothing inside the chain changes — same blocks, same Start block, same logic. What's new is that the workflow is now live and reachable from outside, at a stable, versioned URL. A deployment is a snapshot, so you can promote new versions and roll back.

A call from outside

An external request comes in, hits that address, and flows straight into the Start block — the same entry you tested with in the editor. From there the run is identical: the chain executes and the response goes back to whoever called it. An external call is just a run; the only difference is where it came from.

The same workflow, three surfaces

The one deployed workflow can be reached three ways, and you choose which:

  • API — other systems POST to an endpoint and get the response back.
  • Chat — a hosted chat page anyone can open; each message becomes the workflow's input.
  • MCP — the workflow becomes a tool that other AI agents (like Cursor or Claude) can call.

Same blocks underneath, same Start — just different doors in.

Runs when they run it

Once deployed, the workflow runs on demand — whenever a caller hits it, with no one in the editor. The same process you built is now an operational system.