What's public
Vidext's programmatic surface, honestly scoped — an MCP server plus a machine-readable discovery layer.
Vidext's public programmatic interface is agent-native: a remote MCP server — Vidext Operator — with typed read operations and an operator conversation for planning, building, and editing learning content, plus a public discovery layer that lets agents understand the product before authenticating.
The surface, precisely
| Surface | Status | What it is |
|---|---|---|
| Operator MCP server | Public, authenticated | Typed read and conversational tools over streamable HTTP at /api/mcp/operator. OAuth 2.1. This is the API. |
| Discovery layer | Public, no auth | llms.txt, OpenAPI, server cards, and .well-known metadata describing the product and the MCP contract. See Discovery endpoints. |
| REST CRUD API | Does not exist | There is no general-purpose REST resource API. Integrate over MCP. |
Connect a client (Claude, ChatGPT, Cursor, Codex, or your own) to one URL and authenticate via OAuth. Use typed tools to search knowledge, inspect modules and operations, or read chats. Use consult_operator for read-only model reasoning and ask_operator for write-capable conversations. Long-running work is asynchronous: follow the typed run.nextAction with get_operation.
What you can build
- Assistant integrations — let Claude/ChatGPT/Cursor users drive Vidext conversationally (5-minute setup).
- Autonomous pipelines — generate document-grounded modules from your systems: reference your sources, ask the operator to build, and continue the conversation until the module is ready (recommended flows).
- Content operations — programmatic edits to published modules: text, quizzes, scene structure, media (tool reference).
Everything runs under the caller's identity and permissions — agents get no privileged path. Explicit creation requests plan and build in one flow, builds produce drafts, and publishing remains an explicit human action in the app.
Last updated on