Skip to content

Integration · How-to guide

MCP

Connect an AI assistant to Kokobi Learning with MCP and OAuth.

Connect your AI assistant to Kokobi Learning using the Model Context Protocol (MCP). Your assistant can use tools to read learning data and perform actions your account is allowed to perform.

You need an AI client that supports remote MCP servers, Streamable HTTP, and OAuth, such as a compatible Claude or ChatGPT client. Availability depends on your plan and workspace policy. No API key is needed.

Connect your assistant

Use this MCP server URL in your AI client:

plaintext
https://learning.kokobi.org/api/mcp

Use the platform URL above, not an organization custom domain.

  1. Copy your full MCP server URL, including /api/mcp.
  2. In your AI client, open its connectors, apps, or MCP settings and add a remote server.
  3. Name it Kokobi Learning and paste the URL. Select Streamable HTTP and OAuth if prompted. Do not configure custom headers or paste a token.
  4. Connect the server to start the sign-in flow. Sign in with your Kokobi account, review the requested permissions, and approve only if you trust the client.
  5. Return to your AI client to finish connecting. Enable Kokobi Learning in a conversation.

Authorize each AI client separately using your own account.

Try a first request

Ask your assistant:

List the courses I can access in Kokobi Learning.

Review proposed tool calls before allowing actions that change data.

Permissions

Your client may request these scopes. OAuth does not grant more access than your Kokobi account already has.

ScopeAccess
mcp:readView information your account can access.
mcp:writeMake changes your account is allowed to perform.
offline_accessStay connected without asking you to sign in each time, until access is revoked.

Tool results are shared with your AI provider. Follow your organization's data policies, connect only trusted clients, and never paste passwords or access tokens into a conversation.

Troubleshooting

  • No option to add a server: check that your client supports remote MCP with OAuth and that custom integrations are enabled for your plan and workspace.
  • Cannot reach the server: check that the URL uses your platform host and ends with /api/mcp. Cloud-hosted clients cannot reach a localhost URL.
  • Sign-in expired: reconnect from your AI client's settings.
  • Access denied: check the account you authorized and its organization permissions. Reconnecting does not grant additional access.