VideoLens
Turn a long video into a written report with timestamped evidence. On the Chrome Web Store.

Paste a link or upload a file, choose a report style, and get a standalone HTML, PDF, Markdown or JSON report grounded in speech, sampled frames and on-screen text. Per-step caching and a cost-aware frame budget keep runs cheap. An MCP server exposes the same pipeline to agents.
- Problem
- Transcript summarisers miss what is on screen. Bug reports, lectures and demos carry their evidence in the frames.
- Build
- A multi-signal pipeline that combines speech, sampled frames and OCR before writing, with every claim anchored to a timestamp. Outputs are portable artefacts, not chat replies.
- Outcome
- Working alpha with a public site and an MCP server so coding agents can extract a timestamped repro from a screen recording.
Related work
All work
Live
LiveForward an email. The inbox classifies it, files it and asks only when it must.
- TypeScript
- Supabase
- Claude API
- AgentMail
LiveThe scanner, where the code is written: MCP server, OpenClaw skill and editor extension.
- TypeScript
- MCP SDK
- OpenClaw
- VS Code API