Connect your agent
Add Glass Page to your agent once, and sign in. Then ask it to publish.
The server #
Paste this address into your agent. It takes Streamable HTTP.
Name it glass-page, as the steps do.
Add it in one click #
Each button opens the app, which asks you before it adds Glass Page. Then you sign in in your browser.
Choose your agent #
- Claude On the web or Claude Desktop, as a custom connector.
- Claude Code One command, for every project.
- ChatGPT As a custom MCP server, on Business, Enterprise and Edu.
- Cursor One click, or a line in mcp.json.
- VS Code One click, or one command.
- Codex One command, shared with its IDE extension and the ChatGPT desktop app.
- LM Studio One click, or a line in mcp.json.
- With an API Key For an agent that can't sign you in.
Or let your agent connect itself #
Copy this into any agent. It adds Glass Page the way its app does, and asks you to sign in.
Connect yourself to Glass Page's MCP server, so you can publish HTML pages (Panes) to glasspage.ai and read the feedback people leave on them.
Server: https://mcp.glasspage.ai/mcp (Streamable HTTP). Name it "glass-page".
1. Add it as a remote MCP server the way your harness does: its command for adding MCP servers if it has one, otherwise its MCP config file. If you're not sure how, check your harness's docs or help before editing anything. Add it for my user account, not only this project.
2. Sign in with OAuth if your harness supports it for remote MCP servers. Glass Page registers clients itself, so there's no client ID or secret. Start the sign-in, and I'll finish it in my browser.
3. Only if your harness can't use OAuth: name it "glass-page-key" instead, and have it send an API Key on every request as the header "Authorization: Bearer <key>". I'll create the key at https://glasspage.ai/connections#api-keys. Don't ask me to paste the key into this chat, and never put it in a file this project commits: leave a placeholder or read it from an environment variable, then tell me exactly which file or variable to put it in.
4. Once it's connected (you may need to reload your MCP servers or restart), call list_workspaces to check, and tell me which Workspaces you can see. Don't publish anything yet. Signing in #
With your browser. Most agents sign you in with OAuth: they open your browser, you sign in to Glass Page, and you allow the agent. You don't need a key. Each agent you allow shows in Connected apps, where you can disconnect it.
With an API Key. For a host that can't sign you in, create an API Key in Connections. The agent sends it on
every request, as Authorization: Bearer <key>. Give each
agent its own key, and never commit it to a project.
Either way, your agent acts as you, in every Workspace you belong to.
When an agent asks to connect, you see a page titled Connect an app to Glass Page?, naming the app. Approve it only if you just asked it to connect.