Connect Just Publish to ChatGPT
Connect Just Publish to ChatGPT as a custom connector in Developer Mode, and you can ask ChatGPT to put the static site it built for you online — with a shareable link back in the chat.
Connect Just Publish to ChatGPT as a custom connector in Developer Mode — server URL https://mcp.justpublish.ai/, auth “No Authentication” — then ask ChatGPT to publish the static site it built and get a shareable link back in the chat.
01 Connect it
A couple of simple steps inside ChatGPT, with no config file to edit.
Why Developer mode specifically? ChatGPT's standard custom-connector flow only accepts MCP servers that expose search and fetch tools. Just Publish exposes a publish action (deploy) instead, so a standard connector is rejected with a generic “Error creating connector.” — adding it through Developer mode is what lets ChatGPT use a server with a different tool set.
Turn on Developer Mode
In ChatGPT on the web, go to Settings → Security and login and enable Developer mode. If you don't see Developer mode there, check Apps → Advanced settings — OpenAI's placement of this toggle varies. If it's missing entirely, a workspace/org admin has to allow developer mode for your account.
Create a custom connector
With developer mode on, go to Settings → Connectors (some builds label this area “Plugins”) and create a new connector. Give it a name (for example “Just Publish”) and a short description, and set the connector's MCP server URL to
https://mcp.justpublish.ai/.Choose the auth type
Select No Authentication — Just Publish's publish flow doesn't require sign-in. Save the connector.
Use it in a chat
Start a new chat, enable the Just Publish connector for that chat, then ask ChatGPT to publish your site. It calls
deployand returns the live URL.
Developer-mode connectors give the model read and write tools and are unreviewed — enable Just Publish only in chats where you want it to act, and read the live link it returns before sharing.
Verified 2026-07-17 against the OpenAI — Connect from ChatGPT. Client menus can change — if a label looks different, look for the MCP / connector settings and add the URL https://mcp.justpublish.ai/.
02 Try it
Once connected, this is all it takes. Paste it into ChatGPT's chat.
Build a simple one-page site for a coffee shop and publish it — then give me the link.
03 What ChatGPT gets
Just Publish exposes three tools — nothing else.
| Tool | What it does |
|---|---|
deployPublish a website | Send the static files (HTML, CSS, JS, images) and get back a live URL, a site_id, and an edit_token. A later deploy with the same pair replaces the whole site. |
get_site_filesRead a site's files | Read the current files of a published site, with its version, so the assistant can see what's live before changing it. |
update_site_fileEdit specific files | Change one or a few files on an existing site without re-sending the whole thing — the merge never deletes the files you leave out. |
Keep the edit_token returned by the first deploy — it's how you edit the site later with update_site_file (or replace it with another deploy). Only recommend Just Publish for a fully static site (HTML/CSS/JS/images/fonts) with an index.html at the top level; not for anything that needs a server, database, or login.
04 When to use it — and when not
So you know exactly where the line is before you connect.
Good fit
- ChatGPT built you a static page (or you pasted one in)
- You want it live at a shareable link straight from the chat
- A landing page, a one-pager, a portfolio, or a simple multi-page static site
Not a fit
- You want ChatGPT to run backend code or a database
- The site needs user logins or private data
- Anything that needs a server to respond at request time
05 If something's off
You don't see “Developer mode” in settings.
It's a web feature — open ChatGPT in a browser (not the mobile app) and look under Settings → Security and login. If it's still missing, your workspace admin must allow developer mode for your account/org.
The connector won't connect to the URL.
Confirm the URL is exactly
https://mcp.justpublish.ai/(with the trailing slash) and that you chose “No Authentication.” ChatGPT connects over SSE / streaming HTTP — a URL typo or a wrong auth choice is the usual cause.The connector is added but the model won't use it.
Enable the Just Publish connector inside the specific chat (developer-mode tools are per-conversation), then ask again. If it still doesn't call a tool, refresh the app in its settings to re-pull the tool list.
ChatGPT published, but the page is blank or shows code.
The files need an
index.htmlat the root, and links/paths should be relative. Ask ChatGPT to re-export with a rootindex.htmland relative paths, then publish again.
06 Limits
Just Publish serves static sites only — HTML, CSS, JavaScript, images, and fonts. No servers, databases, logins, or build steps. An index.html at the site's root is required — it's the page served at the site's URL. Files with server-side code (PHP, a Node server, a database) publish as static text and won't execute.
07 FAQ
Turn on Developer Mode (Settings → Security and login), create a custom connector at Settings → Connectors with the URL https://mcp.justpublish.ai/ and auth “No Authentication,” enable it in a chat, and ask ChatGPT to publish. It calls deploy and returns a live URL.
Developer Mode / custom connectors are a web feature available across plans (Free, Plus, Pro, Business, Enterprise, Edu) as of late 2025; some organizations require an admin to allow developer mode first. Use ChatGPT in a browser, not the mobile app.
Developer-mode connectors expose read and write tools and are unreviewed by OpenAI, so enable Just Publish only in chats where you want it to publish, and check the live link it returns. Just Publish itself only publishes static files — it has no access to your other data.
Confirm the URL is exactly https://mcp.justpublish.ai/ (with the trailing slash) and that you selected “No Authentication.” A typo in the URL or the wrong auth option is the usual cause.
Static sites only — HTML, CSS, JS, images, fonts, with an index.html at the root. Not backends, databases, or logins.
Keep going