How to move from OpenAI Agent Builder to FlowRunner

OpenAI shuts down Agent Builder on November 30, 2026. After that you can’t open your workflows, so the code you export before then is the copy you keep.

FlowRunner Prompt to Flow turns that code back into a workflow. Paste one workflow’s Agents SDK export into your AI tool, and it builds the same flow in your workspace, with an approval step that really asks someone. What’s left for you is your OpenAI key, your accounts, and testing.

This is the short version. For a piece-by-piece map, read the Agent Builder shutdown guide. For six of OpenAI’s own templates converted this way, read how to convert Agent Builder workflows to FlowRunner.

1. Export your workflows from Agent Builder

Open each workflow, click Code, choose the Agents SDK tab, pick TypeScript or Python, and copy the code. The ChatKit tab only gives you an embed ID, which points at a workflow that won’t exist after November 30.

• What’s in it: each agent’s instructions, model, and settings, the structured outputs, the branches, and the order everything runs in.
• Web Search blocks it. If any agent has the built-in Web Search tool, the export fails with “Unknown tool type: web_search”. Write down which agents had it, remove the tool, copy the code, and put the tool back.
• What’s left out or broken: in the six templates we exported, human approval came out as a function that always returns true, function tools came out as TODO stubs, uploaded files weren’t in the input, and some exports didn’t run. File search carries only the IDs of your vector stores, which stay in your OpenAI account.

Two more things close on November 30: the Evals dashboard and API (existing evals are already read-only), so save the results you want, and the reusable prompts API (v1/prompts). If your app loads prompts from it, copy that text into your code or the converted agent’s instructions.

2. Connect FlowRunner Prompt to Flow to your AI tool

FlowRunner Prompt to Flow is an MCP server you add to the AI tool you already use. You connect it once and sign in with your FlowRunner account. The server URL is the same for every tool: https://app.flowrunner.ai/mcp.

• Claude (web and desktop): open Settings > Connectors, click + Add, then Add custom connector. Enter FlowRunner and the server URL, click Add, then Connect, and approve the FlowRunner sign-in. On a Team or Enterprise plan, an Owner adds it once under Organization settings > Connectors.
• Claude Code: run these two commands in your project folder, and approve the sign-in page that opens:

claude mcp add --transport http flowrunner https://app.flowrunner.ai/mcp
claude mcp login flowrunner

• ChatGPT (on the web): open ChatGPT Plugins, click Add, then Add custom MCP server. Enter FlowRunner, put the server URL under Connection, choose OAuth, and approve the sign-in. In a chat, type @ and pick FlowRunner.
• Codex: add the server to ~/.codex/config.toml, then run codex mcp login flowrunner:

[mcp_servers.flowrunner]
url = "https://app.flowrunner.ai/mcp"

• Cursor: add the server to ~/.cursor/mcp.json, then open Customize > MCPs and follow the sign-in prompt:

{
  "mcpServers": {
    "flowrunner": { "url": "https://app.flowrunner.ai/mcp" }
  }
}

If you don't have a FlowRunner account yet, the Free plan needs no card, and every new account starts with 14 days of Professional. The setup guide in the docs has each tool step by step.

3. Convert one workflow at a time

Start a new chat for each workflow. Paste this request, then the exported code:

Rebuild this OpenAI Agent Builder workflow as a FlowRunner flow. The input is the Agents SDK code Agent Builder exports, in TypeScript or Python. It describes each agent’s instructions, model, and structured output, and the branches between them. Keep the same behavior. The export is often incomplete: human approval comes out as a function that always returns true, function tools as TODO stubs, uploaded files aren’t in the input, and file search has only vector store IDs. Read the code as a description of the workflow, not as code to run. Before you build, tell me anything you can’t map one to one, and ask me about anything the code doesn’t say.

Then add a line on what the code can’t tell it, such as which agents had Web Search and who approves on which channel. Answer the questions, and the flow appears in your workspace.

4. Finish what the export can’t carry

The builder ends each conversion with a short list. Expect some of these:

• Your OpenAI key. Add it once in the workspace. Every agent runs on it, billed by OpenAI as before.
• Accounts. Connect Gmail or Slack for approvals, and a web search service if your agents searched the web.
• Documents. Upload what your vector stores held to a Knowledge Base and attach it to the agent.
• How the flow starts. In Agent Builder a person typed into a chat. Here it’s an API call from your app, a webhook, a schedule, or an incoming email. Calls from your app carry your workspace key, so make them from a server.
• Approvals. The always-true stub becomes a human-in-the-loop step that reaches the person on email, Slack, WhatsApp, phone, or a Form.
• Forms and agent tools. If the flow needs a form, or an AI agent that calls another flow as its tool, the AI tells you which of these to add in the flow editor.

5. Test, then switch over

Run each flow with the messages you used in Agent Builder’s Preview, or ask your AI tool to run it, and compare the answers. When they match, point your app at the flow and turn the Agent Builder workflow off, a few days before November 30.

How Agent Builder maps to FlowRunner

Agent Builder	FlowRunner	What changes
Workflow	Flow	Starts from a call or an event, not only a chat.
Start (input_as_text)	The flow’s Initial Data	Name as many input fields as you need.
Agent	AI Agent	Same instructions and model, on your own key.
Guardrails	Condition with an AI Question	A yes or no check on the real input.
If/else	Condition, Value Router, AI Router	AI Router branches on meaning.
Human approval	Human-in-the-loop step	It asks a real person and waits.
File search	Knowledge Base	Upload the documents again.
MCP	MCP servers on the agent	Attach a whole server or single tools.
Function tools	An integration action, Custom Cloud Code, or another flow	Built for real, not a stub.
Set state	Set Variables or Shared Memory	Shared Memory lasts across runs.
End	Return Result	Sends the answer back to the caller.
Preview and traces	Run history	Every run block by block, with inputs and outputs.

Quick wins on day one

• Approvals that wait. The request reaches the person on email, Slack, WhatsApp, phone, or a Form and waits up to 30 days on Free, Starter, and Growth, and up to a year on Professional and above. The export’s approval was a function that always returns true.
• Nothing to host. Flows run in FlowRunner’s cloud and start from an API call, a webhook, a schedule, an incoming email, or an event in one of 2,100+ integrations.
• Same OpenAI bill. Agents run on your own key with no markup, and a run counts once however many agents it has.
• A free place to land. The Free plan is $0 with 100 executions a month and no card, with unlimited users and workflows, and every new account starts with 14 days of Professional.