Apr 25
Public API + MCP Server endpoint for programmatic notebook access
**Use case**
AI agents and custom workflows (Claude Code, ChatGPT Desktop with MCP, Cursor, custom Python scripts) increasingly need programmatic read/write access to knowledge bases. Kortex sits on top of NotebookLM and is the natural integration point — but right now there's no documented way to push sources, read notebooks, or trigger actions from outside the Chrome extension.
**Proposal**
Two complementary additions:
1. **REST API with token auth** — minimum useful surface:
- `POST /sources` (notebook_id, type=url|markdown|pdf, content, tags)
- `GET /notebooks` / `GET /notebooks/{id}/sources`
- `POST /notebooks/{id}/audio` (trigger audio overview generation)
- `GET /notebooks/{id}/export` (Markdown ZIP)
2. **MCP server endpoint** (Model Context Protocol, anthropic.com/news/model-context-protocol) — lets every MCP-aware AI client treat Kortex as a tool. This is the emerging standard and would put Kortex ahead of every other NotebookLM extension.
**Impact**
Unlocks Kortex for the entire AI-agent ecosystem: Claude Code, OpenAI Agents SDK, Cursor, Continue, custom CLI tools. Today users like me work around it with `notebooklm-py` (unofficial CLI hacking the NotebookLM frontend) — a public Kortex API would replace that with something stable and supported.
Also a defensive moat: when Google eventually ships a NotebookLM API, Kortex needs to be more than a UI wrapper to retain value. An open API + MCP layer makes Kortex the orchestration tier.
Pending