Connect Coursified to Claude or any MCP client, and it can search the catalogue, read your courses, and build new ones from a video — on your account, with your permission, and without the ability to publish or delete anything.
Coursified turns a video into a structured course: modules, lesson notes, quizzes and an assessment. The MCP connector makes that available to an AI assistant, so instead of switching to Coursified you can ask the assistant directly.
It speaks the Model Context Protocol over Streamable HTTP. Any client that supports remote MCP servers with OAuth can connect — Claude on the web and in Claude Code today, and others as they add support.
There is one URL, and it is the same for everybody. Add it as a custom connector and your client will walk the rest: it discovers the authorization server, asks you to sign in to Coursified, and shows you a consent screen listing exactly what it is about to be allowed to do.
https://mcp.coursified.app/mcpSettings → Connectors → Add custom connector, then paste the URL above. Leave authentication on the detected setting: the server declares OAuth, and Claude will pick it up on its own.
In Claude Code, add it from the terminal instead:
claude mcp add --transport http coursified https://mcp.coursified.app/mcpThe connector follows the standard discovery chain, so a client that implements it needs nothing from us beyond the URL. An unauthenticated request returns a 401 naming where to find the protected-resource metadata, that document names the authorization server, and the server's metadata carries the endpoints, the scopes, and support for both dynamic client registration and Client ID Metadata Documents.
Proof-key exchange (PKCE, S256) is required. There is no client secret to hold.
These are the permissions a client can ask for, in the same words the consent screen uses. A client asks for a subset; you see the list before you approve it, and you can revoke the whole connection later from Settings.
catalog:readSearch and read courses anyone can already see on Coursified.
courses:readIncluding private drafts and their lesson notes.
courses:writeNew courses are private drafts. Only you can publish them.
pathways:writeCreate learning paths and add courses to them.
progress:readHow far through each course you are, and your quiz results.
profile:readYour goals, level and interests, so it can suggest what fits.
offline_accessKeep access after today, without asking you to sign in again. Revoke it any time in Settings.
This is the part worth reading twice, because it is the reason the rest is safe to allow.
There is no scope that publishes a course and no scope that deletes anything — not in the consent screen, not in the server, and not in the tool vocabulary behind it. A course an assistant creates is a private draft. Making it public, and removing anything at all, stays something you do yourself in the Coursified app.
That is enforced where tokens are issued rather than promised here: the scope list the authorization server will mint has no publishing or deleting entry in it, and a test fails the build if one is ever added without this page changing too.
Cannot publish courses publicly, and cannot delete anything.
The connector reads and writes the same things you can see in the app, scoped to your account, and nothing else. It never reads another account's private courses.
Requests are logged for operations. A log line records which tool was called, whether it succeeded and with what error if not, how long it took, whether the caller used OAuth or an API key, which app connected, the permissions it holds, and a one-way hash of the account id. Nothing else — no course content, no prompts, no tokens or keys.
The full picture is in the privacy policy, which covers the product as a whole.
OAuth is the main door and should be used by anything that can open a browser window. A script, a cron job or a CI step cannot, so Coursified also issues scoped API keys.
Create one in Settings → API keys, choose the scopes it should carry, and send it as a bearer token to the same URL. A key carries the same permissions and hits the same checks as an OAuth token — it is a different way to present a credential, not a way around one.
Authorization: Bearer ck_your_key_hereGeneration costs real money, so an agent credential carries its own allowance rather than spending yours: a few courses a day and a rolling monthly ceiling, both counted per credential. Retries count too, at a fraction of the cost.
When a limit is reached you get a typed refusal saying which one and, where waiting would help, how long to wait. Revoking a key takes its spending with it.
Something broken, something confusing, or a tool that behaved differently from this page? Write to us — a description of what you asked the assistant to do is usually enough to find it.
hello@coursified.app