Connect ChatGPT or Claude to Studeia (MCP server)

How it works

Studeia implements the Model Context Protocol (MCP) with Streamable HTTP transport and OAuth 2.1 (dynamic client registration, PKCE S256, rotating refresh token). The AI assistant connects with your account, not with an institution key: everything it does goes through the same permission rules as the web interface and is recorded in the audit log.

	Institutional Administrator	Teacher
Read	Everything in the tenant: courses, lessons, students, classes, grades, reports, settings	Courses they create/collaborate on/teach; students in their classes
Write	Courses, modules, lessons, teacher guides, classes, enrollments, settings	Courses and lessons in their own courses
Out of scope	Billing, SSO/SCIM, provider keys, the connector toggle itself	same

1. Enable at the institution level (admin)

Go to Settings → AI Connectors (ChatGPT / Claude).
Turn on Allow MCP connections. Optionally disable Teachers can also connect.
Copy the MCP server URL — typically https://<slug>.studeia.com/api/mcp (or your verified custom domain).

Disabling the connector immediately revokes all active connections.

2. Connect in Claude

Claude.ai / Desktop / mobile

Settings → Connectors → Add custom connector.
Paste the MCP server URL. Leave Client ID/Secret blank (automatic registration).
Click Connect: you will be taken to the Studeia login (if not already logged in) and the consent screen.
Review the permissions (read, write, 30 days) and the privacy notice → Authorize.

Claude Code

claude mcp add --transport http studeia https://<slug>.studeia.com/api/mcp

Claude Code opens the browser for consent and receives the callback at http://localhost:<port>/callback.

3. Connect in ChatGPT

Settings → Connectors → Advanced → enable Developer Mode.
Create → name (e.g., "Studeia") → MCP server URL → authentication OAuth → Create.
Complete the Studeia login and consent flow.
In the chat, enable the connector under Tools. Write actions require manual confirmation in ChatGPT.

The search and fetch tools follow the ChatGPT contract, so the connector also works in deep research.

4. What to ask the assistant

"Generate a report for class 3rd Grade A showing at-risk students and the skills with the lowest mastery."
"List the lessons in the Digital Skills course that are missing a teacher guide and create one for lesson 4."
"Rewrite the slides for the Digital Vault lesson in simpler language and publish them."
"Which users have never accessed the platform? Export as a table."
"Show the gradebook API endpoints and submit grade 8 for student X on assessment Y."

Available tools (summary): tenant overview and settings, courses and modules (list, create, edit, publish, reorder), lessons (full content, create, edit validated content by type, teacher guide, delete), users, classes, enrollments, student/course/tutor effectiveness reports, audit logs, search/fetch, and list_api_endpoints + call_institution_api for the rest of the institutional API.

5. Revoke and audit

User: Profile → AI Connections → Revoke.
Admin: Settings → AI Connectors → connections list → Revoke (or disable the connector).
Audit: Logs → mcp.* actions (consent, tokens, each tool call with tool name, ids, and result — never the content).

Privacy and limits

Data queried by the AI (including data of minor students) is processed by the user's account provider (OpenAI/Anthropic). The institution is responsible for legitimate use — the consent screen makes this explicit.
Access token: 1 hour. Refresh token: rotating, maximum 30 days per consent (re-authorization required afterward).
Rate limit per connection and per IP on OAuth endpoints.
Not available via MCP: file uploads, billing, SSO/SCIM, AI/video provider keys, and the connector configuration itself.

For developers

Discovery: /.well-known/oauth-protected-resource/api/mcp and /.well-known/oauth-authorization-server (relative to host).
POST /api/mcp without a token responds with 401 and WWW-Authenticate: Bearer resource_metadata=….
Dynamic registration only accepts callbacks from Claude, ChatGPT, and loopback (http://localhost:<port>/callback).
Scopes: read, write, offline_access.

FAQ

Does Studeia have an MCP server for ChatGPT and Claude?

Yes. Each institution exposes a remote MCP server at https://<slug>.studeia.com/api/mcp (Streamable HTTP + OAuth 2.1 with dynamic client registration). It works with Claude.ai, Claude Desktop, Claude mobile, Claude Code, and ChatGPT's custom connectors (Developer Mode).

What can the AI do once connected?

Exactly what the connected user can do on the web: read courses, lessons (including full content and teacher guides), classes, enrollments, student and course reports, and — with write permission — create and edit courses, modules, lessons, teacher guides, and institution settings. There are ~30 dedicated tools plus a generic tool that reaches the entire documented institutional API.

Can a connected teacher see data from the entire school?

No. The MCP token acts as the real user: teachers can only view and edit the courses they created, collaborate on, or teach (classes), and reports only for students in their classes. Institutional administrators see everything in the tenant. Global platform administrators cannot connect.

How do I revoke the AI's access?

In the user profile (AI Connections) or, for administrators, in Settings → AI Connectors (list of all tenant connections). Disabling the connector revokes all connections immediately. Access tokens last 1 hour; the refresh token lasts at most 30 days and is rotated.

Does student data go to OpenAI or Anthropic?

Data queried by the assistant (including data of minor students) is processed by the AI provider of the user's account. The consent screen explicitly warns about this, and every tool call is recorded in the institution's audit logs. Use only for legitimate educational purposes and comply with applicable data protection regulations.