Add SpringRoll to ChatGPT as a custom MCP server
SpringRoll has an endpoint made for ChatGPT, separate from the one coding agents use. You add it from Plugins with three values. There is no API key to paste: ChatGPT identifies itself to SpringRoll's sign-in and opens it in your browser.
- Open Plugins in ChatGPT, choose the plus button, then Add custom MCP server.
- Name: SpringRoll. You can leave the description empty.
- MCP server URL: https://springroll.dev/api/mcp/chatgpt
- Authentication: OAuth.
- Read the risk notice, confirm it and choose Create as a plugin. ChatGPT connects and lists SpringRoll's tools.
- Install the plugin and authorize. Choose Connect SpringRoll to start as a guest without a SpringRoll account or sign-in. If already signed in, use your existing workspace.
- Start a new chat, type @ and choose SpringRoll.
Name: SpringRoll
MCP server URL: https://springroll.dev/api/mcp/chatgpt
Authentication: OAuthDo you still need ChatGPT developer mode?
Not in the current Plugins flow. OpenAI's documentation now adds a custom MCP server from Plugins, with no separate developer mode switch. Guides written in late 2025 and the first half of 2026 describe turning on Developer mode under Settings, Connectors and Advanced settings, then choosing Create. If your workspace still shows that route, enter the same three values: the name SpringRoll, the URL above and OAuth.
Either way, ChatGPT treats SpringRoll as a full MCP server that can change things, not a read-only search connector. That is why SpringRoll asks you to confirm every change itself, as the next sections show.
Deploy an app from ChatGPT
ChatGPT has no project folder of its own, so give it the source: a repository SpringRoll can read, a project archive you upload to the chat, or the files ChatGPT has written in the conversation. Then ask it to deploy:
Prepare this project for SpringRoll. Read the Company Blueprint, inspect my source and discover supported workloads and services. Check the build, workspace, owner and deployment workflow; show the exact deployment proposal for confirmation, then follow the build to its actual result.What happens when ChatGPT deploys through SpringRoll
ChatGPT reads your workspace and your Company Blueprint, checks which Vercel and Cloudflare workloads fit the app, then calls springroll_prepare_deploy. SpringRoll does not build anything yet. It returns a proposal with a preview of exactly what will be deployed and where, and a link to confirm it.
Apps on the Direct workflow go to Production once you confirm. Apps on the Staged workflow start in Development and need a configured reviewer's approval to reach Production.
Confirm the deployment in SpringRoll
Every change ChatGPT prepares through SpringRoll, whether a deployment, a promotion, a rollback or a record update, comes back as a proposal with a confirmation link. Open the link, check the proposal in SpringRoll and confirm it. SpringRoll then runs that exact proposal once. A proposal expires, and a changed proposal needs a new confirmation.
If you would rather confirm in the chat, grant the separate springroll:confirm permission when you connect. ChatGPT may then submit a proposal only after you approve that exact, current proposal in the conversation, and SpringRoll records your approval note with it. An existing connection never gains this permission by itself, and reviewer approvals your company requires still need that reviewer.
Watch the build and open the app in ChatGPT
After you confirm, ask ChatGPT to show the deployment. springroll_panel opens a live card with the release, its status and the latest build log lines. The card refreshes while it is on screen and stops when the build is READY or fails.
When the app is READY, ask ChatGPT to preview it. springroll_preview shows it in a desktop or mobile frame, or gives you the checked Open app link when the app cannot be embedded. If the build fails, ask ChatGPT to read the logs; with manager access it can read the failed release's source, propose a fix and prepare a new deployment for you to confirm.
What SpringRoll adds to ChatGPT
These are the SpringRoll tools ChatGPT uses most when shipping and running an app.
Scroll sideways to see all columns. With the table focused, use the arrow keys.
| Tool | What it does |
|---|---|
| springroll_context, springroll_profile | Shows which organization, identity and permissions the connection acts as. |
| springroll_blueprint, springroll_workloads | Reads your company's published building guidance, and the Vercel and Cloudflare workloads and services an app can use. |
| springroll_prepare_deploy | Prepares a deployment from a repository, files written in the chat, a project archive or a file you upload to ChatGPT. |
| springroll_deployment, springroll_deployments, springroll_panel | Reads a deployment's status, release and redacted logs, and shows them in a live card. |
| springroll_preview | Shows the running app in a desktop or mobile preview, or the checked Open app link when it cannot be embedded. |
| springroll_prepare_promote, springroll_prepare_rollback, springroll_prepare_cancel | Prepares a promotion, a restore of an earlier release or a cancellation, each for you to confirm. |
| springroll_repair_context, springroll_source_file | Reads a failed release's source and redacted logs so ChatGPT can propose a fix. Manager access is required. |
| springroll_apps, springroll_app | Finds the apps you can open and reads an app's record and deployment history. |
| springroll_prepare_security_audit, springroll_security_audit | Prepares a security audit of a release under your company's policy and reads its findings. |
| springroll_data_products | Lists governed, read-only company data available through SpringRoll Connect. |
| springroll_operation | Checks whether a confirmed action ran. An unresolved change is reconciled, never blindly repeated. |
What ChatGPT cannot do through SpringRoll
Connecting ChatGPT does not let it bypass your team's rules.
- It cannot deploy, promote, roll back, suspend or retire anything until you confirm the exact proposal.
- It cannot record an approval your company requires from a reviewer. Those decisions stay with people.
- It never receives a credential, connection string or secret, and it cannot grant itself access to company data.
- It sees only the apps and deployments your SpringRoll role allows, and every change is recorded in your organization's audit log.
SpringRoll in ChatGPT Business and Enterprise workspaces
In a company workspace, ChatGPT's admins decide whether members can add custom MCP servers and which actions a connection may take. OpenAI's documentation says account and workspace policies apply to adding and using custom MCP servers, so if Add custom MCP server is missing from Plugins, ask your ChatGPT admin.
SpringRoll's own permissions apply on top. The person who authorizes the connection acts with only their SpringRoll role, in the workspace they chose at sign-in, and SpringRoll's Direct or Staged workflow decides what still needs a reviewer.
Sign-in: what SpringRoll asks ChatGPT for
SpringRoll's ChatGPT endpoint follows the MCP authorization specification. ChatGPT reads SpringRoll's discovery document, identifies itself with a published client metadata document or by registering, and asks only for the permissions its tools use. Grants made for coding agents at SpringRoll's other endpoint cannot be used here.
curl https://springroll.dev/.well-known/oauth-protected-resource/api/mcp/chatgpt
# {"resource":"https://springroll.dev/api/mcp/chatgpt", "authorization_servers":[...], "scopes_supported":[...]}Scroll sideways to see all columns. With the table focused, use the arrow keys.
| Permission | What it allows |
|---|---|
| springroll:read | Read your workspace, apps, deployments, logs, security findings and governed data products. |
| springroll:deploy | Prepare deployments, promotions, restores, cancellations, suspensions and retirements for you to confirm. |
| springroll:write | Prepare security audits and changes to app records and connected accounts for you to confirm. |
| springroll:confirm | Optional. Submit a proposal you approved in the chat. Consented separately and never added by a refresh. |
Troubleshooting SpringRoll in ChatGPT
Match what you see.
- Add custom MCP server is missing: your plan or workspace policy does not allow custom servers yet. Ask your ChatGPT admin.
- ChatGPT shows different tools from this guide: you added the coding-agent endpoint. Remove it and add https://springroll.dev/api/mcp/chatgpt instead.
- ChatGPT still shows old tools after a SpringRoll update: open SpringRoll in Plugins, choose Refresh, then start a new chat.
- You signed in to the wrong workspace: remove SpringRoll in Plugins, add it again and choose the right workspace at sign-in.
- A proposal has expired or changed: ask ChatGPT to prepare it again and confirm the new preview.
- springroll_context reports that no runtime can take a deployment: deployments are not set up for that workspace yet. Ask a workspace administrator.
- A build fails: ask ChatGPT to show the deployment with its logs, fix the cause, then prepare a new deployment.
What this guide was checked against
ChatGPT steps were checked against OpenAI's developer documentation for connecting a custom MCP server on 8 November 2026. OpenAI changes this interface often; if a label differs, the three values stay the same: the name SpringRoll, the ChatGPT endpoint URL and OAuth. SpringRoll's tools and permissions come from its live ChatGPT endpoint on the same day.
Frequently asked questions
Do I need the Apps SDK to use SpringRoll in ChatGPT?
No. The Apps SDK is for building your own ChatGPT app. SpringRoll is already an MCP server with its own live cards, so you only add its URL.
Is SpringRoll's ChatGPT endpoint the same as its general coding-client endpoint?
No. ChatGPT and the current packaged SpringRoll skill use https://springroll.dev/api/mcp/chatgpt, with tools built around proposals and confirmation. Manual coding-client setups can use https://springroll.dev/api/mcp for the general tools. Connect the endpoint that matches the guide or installed skill.
Can ChatGPT deploy to Production without me?
No. ChatGPT prepares a proposal and you confirm it. On the Staged workflow a configured reviewer also has to approve Production.
Which ChatGPT plans can add SpringRoll?
OpenAI decides which plans and workspaces can add custom MCP servers, and that has changed during the rollout. If Add custom MCP server appears in Plugins, you can add SpringRoll; in a company workspace your admin controls it.