Paste a YouTube link into Claude or ChatGPT and ask what happens in the video, and you usually get an apology or a guess based on the title. Neither assistant can fetch and watch a video from a link on its own.
An MCP server fixes that. MCP, the Model Context Protocol, is the standard both apps use to call outside tools. Add a server that analyses video, and the assistant can send it the link and read back what was said and shown.
This guide uses OpenCorpo's MCP server, which watches the video itself rather than scraping captions.
Why most YouTube MCP servers break
Most YouTube MCP servers download the video's captions, then let the assistant read them. Two problems follow:
- No captions, no answer. Plenty of videos have no subtitles, or only auto-generated ones in another language. And captions never describe what is shown: slides, products, on-screen text, actions.
- Blocked from servers. YouTube blocks most automated transcript requests from cloud and data-centre networks, which is why many self-hosted transcript tools ask you to buy residential proxies or run them on your own computer.
OpenCorpo takes a different route: the link is handed to Google's Gemini model, which fetches the public video on its side and analyses the picture and the sound together. You get timestamps in MM:SS and the transcript in the language spoken.
What you get back
One analysis returns any of four outputs, for the same price:
| Output | What it is |
|---|---|
| Summary | One paragraph on what the video is and what happens |
| Description | Scene by scene, with start and end timestamps |
| Transcript | Speech line by line, timestamped, with speakers where possible |
| Prompt | A text-to-video brief, shot by shot, for Veo, Sora, Runway or Kling |
Long transcripts are returned in pages, so a 20-minute talk does not flood the conversation.
Step 1: add the connector
Claude (web, desktop or mobile): open Customize, then Connectors, choose Add custom connector, and paste:
https://api.opencorpo.com/mcpChatGPT: in the settings, add a custom MCP server with the same URL and choose OAuth. Custom MCP servers are available on ChatGPT plans that allow them.
Claude Code:
claude mcp add --transport http opencorpo https://api.opencorpo.com/mcp
# then run /mcp inside Claude Code to sign inCursor, VS Code and the Claude and OpenAI APIs take an API key instead: see the setup snippets.
Step 2: sign in and approve
The app opens app.opencorpo.com. Sign in, or create a free account, and approve the connection. The consent page names the app and shows where you will be sent back. No password or API key is ever pasted into the chat, and you can disconnect the app at any time under API keys → Connected apps.
Step 3: paste a link and ask
Turn the connector on in a chat and ask, for example:
- "Summarize this video and list the key moments with timestamps: https://www.youtube.com/watch?v=…"
- "Give me the full transcript of this talk."
- "Write a Veo prompt that recreates the look of this clip, shot by shot."
The assistant calls the analyze_video tool. Short videos come back within the same answer; for longer ones it starts a job and fetches the result with get_video_job a minute or two later.
Prices and limits
- Free tries: 3 analyses per account, each covering the first 5 minutes.
- Then: €0.05 per started minute of analysed video, once per video, whatever outputs you ask for. The quality model costs ×3.
- Length: up to 20 minutes analysed per video. Files up to 500 MB through an upload link.
- Public videos only: private and unlisted videos that need a login cannot be fetched.
Failed analyses are never charged, and the assistant is told how much to top up when the balance runs out.
The same server does two more things
The connector also marks AI-generated images for EU AI Act Article 50 (a signed C2PA manifest, an invisible watermark and a fingerprint) and removes EXIF, GPS and hidden characters from photos, videos and text without re-encoding them. Each has its own free tries. See the MCP server page for every tool and price.
Frequently asked questions
Can Claude watch YouTube videos?
Not by itself: Claude cannot fetch a YouTube video from a link. With an MCP server such as OpenCorpo's added as a connector, it can send the link to a video model and get back a summary, a timestamped scene description and the transcript.
Can ChatGPT summarize a YouTube video?
Yes, once a custom MCP server that analyses video is added. Paste the link and ask for a summary or the key moments. You can also attach a video file in the chat.
What is the YouTube MCP server URL?
For OpenCorpo it is https://api.opencorpo.com/mcp, one remote server for video analysis, AI image marking and metadata removal.
Does it work on videos without captions?
Yes. The model watches the frames and listens to the audio, so it does not depend on YouTube's captions. The video has to be public.
Is it free?
Every account gets 3 free analyses, each up to the first 5 minutes of a video. After that it costs €0.05 per started minute, once per video, from a prepaid balance.