Skip to main content
Connect the ChatGPT work app to GooseWorks so it can use your GooseWorks tools — files, chat history, automations, and the full Ad Remixer — directly in your chats, in your own account and billed to your own credits.
This is the same gooseworks MCP server documented in MCP Servers and Ad Remixer MCP. ChatGPT work connects with a token header: Authorization: Bearer cal_.... Same tools, same account.

​
Before you start

  • Custom MCP servers are added under Settings → Plugins → MCPs.
  • You’ll need a user-scoped GooseWorks MCP token with the mcp:access scope. The first step below shows exactly where to create it.

​
Connect

1

Create your GooseWorks MCP token

In GooseWorks, go to Settings → Coding Agents and click Create token under Your tokens. You can also open Manual MCP config and click Create MCP Token — both create the same user-scoped MCP token.
GooseWorks Settings → Coding Agents page with the Create token button under Your tokens
The new token is shown once — copy the cal_… value now (it won’t be shown again). You’ll paste it into ChatGPT in the next step.
GooseWorks 'Token created' box showing a cal_ token to copy
2

Open the custom MCP form

In ChatGPT, open Settings → Plugins, select the MCPs tab, and click + Add server. Fill in the Connect to a custom MCP form:
  • Name: Gooseworks
  • Type: Streamable HTTP
  • URL:
Leave Bearer token env var empty.Under Headers, add:
  • Key: Authorization
  • Value: Bearer cal_your_token_here
Replace cal_your_token_here with the token you just copied, then click Save.
ChatGPT custom MCP form for GooseWorks: URL set to https://mcp.gooseworks.ai/mcp, Bearer token env var left empty, and an Authorization header with a Bearer cal_ token value
The Bearer token env var field is only for the name of an environment variable, such as GOOSEWORKS_MCP_TOKEN. Don’t paste Bearer cal_... into that field. If you prefer the environment-variable path, set the environment variable to the raw cal_... token and enter only the variable name in the field.
3

Confirm it's connected

Back on the MCPs tab, gooseworks appears in the Servers list with its toggle switched on. The GooseWorks tools are now available in your ChatGPT chats.
ChatGPT Plugins → MCPs tab showing the gooseworks server enabled in the Servers list

​
What you can do

Once connected, ChatGPT can, on your behalf and in your account:
  • Make, edit & analyze ads — research a brand, remix templates into branded creatives, re-roll and edit them, and diagnose campaign performance. Generation runs through the same cloud workflow as the web app — billed to your credits.
  • Work with your files, chat history & automations — the standard MCP Servers tools.
  • Discover & load skills — ask ChatGPT to find and use any GooseWorks skill.
Static image ads work here; video ads render locally. Static ads are generated server-side, so they work over this connector. Video ad rendering runs on the machine executing the agent (it needs network egress + a local ffmpeg toolchain), so it works only in local Claude Code or the GooseWorks desktop app — not inside the ChatGPT work sandbox, which blocks the required network access. From here you can still set up a video project (brand, template, script); ask GooseWorks for a static ad to finish in-app, or render the video locally.

​
Managing the connection

  • Access — the token is scoped to your account. It can use the GooseWorks tools and spend ad credits; it cannot change your account settings.
  • Approvals — ChatGPT asks you to confirm write actions (anything that creates or changes data) before it runs them. Read-only tools run without a prompt.
  • Disconnect — toggle the server off, or remove it, on the MCPs tab at any time. To rotate the token, create a new MCP token in GooseWorks, update the Authorization header value in ChatGPT work, then revoke the old token.

​
Troubleshooting

  • Server won’t enable / “couldn’t connect” — double-check the URL is https://mcp.gooseworks.ai/mcp and that the Headers section includes Authorization with a value like Bearer cal_.... Retry once; the first request can occasionally time out.
  • Unauthorized / missing Bearer token — make sure you didn’t paste the token into Bearer token env var. That field expects an environment variable name, not the token value.
  • Some workspaces or ad projects are missing — create the token from Manual MCP config so it is user-scoped. User-scoped tokens let the MCP resolve the right workspace or organization per request.
  • No MCPs tab or “Add server” button — make sure you’re on the latest ChatGPT version; the Plugins → MCPs panel is where custom MCP servers are added.
  • Only some tools show up — ChatGPT limits how many tools a connector exposes at once. The core Ad Remixer and skill tools are prioritized; ask ChatGPT to “list gooseworks tools” to see what’s available.

​
Other clients

  • Claude Code, Cursor, Codex — also a token header; run npx gooseworks install or see MCP Servers.
  • Claude (claude.ai and Cowork) — connects by signing in (OAuth) instead of a token; see Claude.
It’s the same server and the same account either way.