Model Context Protocol
Cubite ships an MCP server, so an AI assistant can operate your learning site directly. Build a course from a syllabus, enroll a cohort, upload a SCORM package, restyle the homepage, publish a blog post, all from chat. Every connection is scoped to one site and to the permissions you approve, and you can revoke it at any time.
Your server URL
https://cubite.io/api/mcp/mcpOne URL for every site. You choose which of your sites to connect when you sign in.
The server exposes 54 tools. The assistant only sees the ones your connection has permission for, so a read-only connection genuinely cannot write.
Create courses, write lessons and units, add quizzes, assignments, and graded activities, and upload SCORM or xAPI packages.
Invite and create users, enroll individuals or a whole cohort, manage groups, and revoke access.
Build multi-course journeys, order the steps, set prerequisites, and enroll learners into the path.
Update settings and theme, rebuild the homepage, and publish pages and blog posts.
Claude and ChatGPT sign in through your browser, so there is no key to copy. Command line and server-to-server clients use a scoped API key instead.
claude mcp add --transport http cubite https://cubite.io/api/mcp/mcpRun /mcp in Claude Code to complete the browser sign-in on first use.
OpenAI has moved these menus more than once. If the wording does not match, look for connectors or plugins in settings.
Create a scoped API key in Admin, then your site, then Integrations, and send it as a bearer header. The key is shown once.
{
"mcpServers": {
"cubite": {
"httpUrl": "https://cubite.io/api/mcp/mcp",
"headers": {
"Authorization": "Bearer ck_your_key_here"
}
}
}
}Any client that can send Authorization: Bearer ck_... works against the same URL. Cursor, Windsurf, and custom agents all use this shape.
Permissions are granted per connection. The OAuth screen lists exactly what is being requested before you approve, and an API key carries only the scopes you gave it. A write permission also grants the matching read.
| Permission | Grants |
|---|---|
| pages:read / pages:write | Read, create, and edit pages and blog posts |
| courses:read / courses:write | Read, create, and edit courses and their content |
| enrollments:read / enrollments:write | Read and manage course enrollments |
| users:read / users:write | Read, create, and invite users |
| members:read / members:write | Read and manage site members and their roles |
| groups:read / groups:write | Read and manage groups and cohorts |
| learning-paths:read / learning-paths:write | Read and manage learning paths |
| question-banks:read / question-banks:write | Read and manage question banks |
| packages:read / packages:write | Read and upload xAPI and SCORM packages |
| site:read / site:write | Read and update site settings, theme, and homepage |
| media:write | Upload images and files |
Grant deliberately. Reading users, members, or enrollments exposes learner names and email addresses to the assistant, which means to that assistant vendor. Writing users can create accounts on your site, and writing packages uploads content that runs in a browser on your domain. If you only want help writing content, grant pages and courses and nothing else.
See our Privacy Policy, Terms, and Security Policy.
The Model Context Protocol is an open standard for connecting AI assistants to external systems. An MCP server exposes a set of tools an assistant can call. Cubite runs one, so an assistant can operate your learning site directly instead of you clicking through the admin.
Anything that speaks MCP over HTTP. That includes Claude (web, desktop, and Claude Code), ChatGPT in developer mode, Gemini CLI, Cursor, and custom agents built on the Anthropic or OpenAI APIs.
No. Every connection is bound to exactly one site, chosen by you when you sign in. There is no parameter an assistant can pass to reach another site, and the server resolves the site from your credential on every single request.
No. Those clients use OAuth: they open a Cubite sign-in in your browser, you pick the site and approve the permissions, and no secret is ever copied or stored by you. API keys exist for command-line and server-to-server clients that cannot open a browser.
Open Admin, then your site, then Integrations. Revoking the connection there immediately invalidates its access. Removing the connector inside Claude or ChatGPT stops that client from calling Cubite, but revoking in Cubite is what actually kills the credential.
No. The MCP server is included with your Cubite site. You bring your own assistant subscription.
Create one, then connect it to your assistant and build your first course by describing it in chat.
Send us a message and we'll get back to you shortly.