How to add an MCP server to ChatGPT with developer mode
Turn on ChatGPT developer mode, create an app for the server's URL, and sign in with OAuth. The steps, with Sume's hosted MCP server as the example.

To add an MCP server to ChatGPT, turn on developer mode under Settings → Security and login, go to ChatGPT Plugins, select the plus button, and create a developer-mode app for the server's remote URL. ChatGPT connects over SSE or streaming HTTP, with OAuth, no authentication, or mixed authentication. For Sume, the URL is https://mcp.sume.com/mcp and the sign-in is OAuth.
ChatGPT's steps come from OpenAI's ChatGPT Developer mode page; its menu names were read on 2026-09-27 and may change. Sume's side comes from the MCP quickstart and OAuth and API keys docs. Sume has no official connector for ChatGPT: this is a developer-mode app pointed at Sume's hosted MCP server.
Does ChatGPT support MCP?
Yes, in developer mode. OpenAI says developer mode provides full MCP client support for all tools, both read and write, and that any tools an app exposes, write actions included, are available subject to confirmation settings. Developer mode is available to Pro, Plus, Business, Enterprise, and Education accounts on the web. OpenAI labels it "Elevated risk": powerful but dangerous, and intended for developers who understand how to safely configure and test apps.
How do I add the server?
- Open Settings → Security and login and turn on Developer mode. The plus button creates developer-mode apps only after that.
- Go to ChatGPT Plugins, select the plus button, and create a developer-mode app with the URL
https://mcp.sume.com/mcpand OAuth. - Sign in on Sume's consent page, which lives on
mcp.sume.com. Read is locked on and Write is off by default; turn Write on if ChatGPT should generate or edit, then continue. - Find the app under "Drafts" in app settings. Its details page toggles tools on or off and refreshes tools and descriptions from the server.
| Setting | What ChatGPT supports | What Sume uses |
|---|---|---|
| Transport | SSE and streaming HTTP | Streamable HTTP at https://mcp.sume.com/mcp |
| Authentication | OAuth, No Authentication, Mixed Authentication | OAuth, with sign-in and consent on mcp.sume.com |
| OAuth client | Static credentials, Client ID Metadata Documents, or DCR when configured | Advertises a dynamic client registration endpoint and PKCE S256 |
| Permissions | Write actions need confirmation by default | Read locked on; Write toggle off by default |
Which authentication should I choose?
OAuth. In Sume's documented flow, the client connects to https://mcp.sume.com/mcp, Sume returns an OAuth challenge, and you sign in and consent on the MCP host. Sume's docs describe that flow for MCP clients in general, not ChatGPT's screens. Sume's other option, an API key in an Authorization: Bearer or x-api-key header, is not among the authentication types OpenAI lists for developer mode, and Sume's quickstart says to prefer OAuth for interactive clients and not to paste API keys into chat.
With OAuth, Sume's consent grants mcp:read, plus mcp:write when you turn Write on; there is no mcp:paid scope. In current code, an access token lasts one hour and the server issues no refresh token. Sume's MCP OAuth flow walks through discovery, consent, and PKCE.
Why does ChatGPT ask me to confirm Sume's tools?
Because they are marked as writes. OpenAI says write actions require confirmation by default, ChatGPT respects the readOnlyHint tool annotation, and tools without it are treated as write actions. Sume's server currently sets readOnlyHint: true on read tools such as tools_list and jobs_wait, and false on every write and paid tool, so generate_video asks for approval. You can remember an approve or deny choice for a tool for the rest of a conversation; new conversations and refreshes prompt again. MCP tool annotations explains the hints.
How do I use the tools in a chat?
Choose Developer mode from the Plus menu and select the app for the conversation. OpenAI suggests naming the app and the tool, so a first check could be: "Use the Sume app's tools_list tool and summarize the tools." Before any paid call, ask for dry_run=true, which previews the cost without submitting the job; paid calls also need an idempotency_key.
What are the risks and limits?
- OpenAI warns about prompt injections, model mistakes on write actions that could destroy data, and malicious MCP servers that try to steal information.
- Hosted MCP cannot read files from your laptop.
- Sume's basics page says hosted MCP still works but is not the primary path today. A product integration calls the Developer API or a Format.
Sources
Related posts
More in Integrations
- How to add subtitles to a video in Python
Add subtitles to a video in Python with Requests: POST the video URL to Sume's /v1/video-captions, poll the job, then read the captioned video_url.
- Airflow HTTP sensor: wait for an AI video job to finish
Submit an AI video job with Airflow's HttpOperator, then wait with an HttpSensor in reschedule mode that passes once the job's status is completed.
- Airtable automation video generation API: a video per record
Use an Airtable Run a script action to call POST /v1/videos with a callback_url, then catch Sume's webhook in a second automation and save the URL.
- Amazon Q MCP server: add Sume's hosted MCP in the IDE
Amazon Q Developer in the IDE takes HTTP MCP servers. Add Sume's hosted MCP with an API-key header or OAuth, then set its paid tools to Ask.
Written by Sume