New: try each tool in your browser, no account needed. Then one API key with 3 free tries of every API. Get your key →
OpenCorpo

One MCP server for YouTube video analysis, C2PA image marking and metadata removal

Paste one URL into Claude, ChatGPT or Cursor. Then ask it to summarize a YouTube video, mark an AI image for EU AI Act Article 50, or strip the GPS data from a photo, and it calls OpenCorpo on your account.

MCP server URL

https://api.opencorpo.com/mcp

Streamable HTTP. Claude and ChatGPT sign in with your OpenCorpo account; developer tools use your API key.

Create your free account →Set it up in your app
  • Remote: nothing to install, no proxy, works on the web and on mobile
  • 3 free tries on each tool, then pay per call
  • The same account, balance and prices as the API

Setup

Add it to Claude, ChatGPT, Cursor, VS Code and Claude Code

Chat apps sign in with your OpenCorpo account: nothing to copy. Developer tools and APIs take your API key.

Claude (web, desktop and mobile)

  1. Open Customize, then Connectors, and choose Add custom connector.
  2. Paste https://api.opencorpo.com/mcp and connect.
  3. Sign in to OpenCorpo and approve. Then turn the connector on in a chat from the + menu.

ChatGPT

  1. In ChatGPT's settings, add a custom MCP server (available on plans that allow custom connectors).
  2. Enter https://api.opencorpo.com/mcp and choose OAuth.
  3. Sign in to OpenCorpo and approve, then select OpenCorpo in a chat.

The consent page shows which app you are connecting and where you will be sent back. Disconnect any app in your dashboard under API keys, Connected apps.

Claude Code, Cursor, VS Code, the Claude API and the OpenAI Responses API

# Sign in with your OpenCorpo account (opens your browser once)
claude mcp add --transport http opencorpo https://api.opencorpo.com/mcp
# then, inside Claude Code: /mcp  ->  opencorpo  ->  Authenticate

# Or use an API key
claude mcp add --transport http opencorpo https://api.opencorpo.com/mcp \
  --header "Authorization: Bearer $OPENCORPO_API_KEY"

Create a key in the dashboard under API keys: the same key works for the REST API and the MCP server. Full reference in the API docs.

Video

Let Claude and ChatGPT watch YouTube videos

Most YouTube MCP servers only fetch captions. OpenCorpo hands the link to a model that watches the frames and listens to the audio, so your assistant gets a summary, a scene-by-scene description with MM:SS timestamps, the transcript in the original language and a shot-by-shot prompt for Veo, Sora, Runway or Kling, even when the video has no subtitles.

It works with public YouTube links and with video files. Long videos run as a job: the assistant starts it, and gets the result a minute or two later, with long transcripts in pages so they fit in the conversation.

Tools: analyze_video, get_video_job. Try the video summarizer free on the web.

Ask, for example

  • “Summarize this video and give me the key moments with timestamps: https://www.youtube.com/watch?v=…”
  • “Write a Veo prompt that recreates the look of this clip, shot by shot.”

AI image marking

Sign and verify C2PA Content Credentials from your agent

Before an AI-generated image is published, your agent can mark it the way EU AI Act Article 50 asks: a signed C2PA manifest, an invisible watermark that survives the manifest being stripped, and a perceptual fingerprint in a registry. It gets back a link to the signed PNG.

It can also check any image for the three layers and say whether it found all, some or none of them. Our signing certificate is self-issued for now, so third-party C2PA validators show the signature as untrusted; the watermark and fingerprint are checked by OpenCorpo.

Tools: sign_image, verify_image. Try the AI image marking free on the web.

Ask, for example

  • “Mark this AI image for the EU AI Act before I publish it: https://…/banner.png”
  • “Was this image marked as AI-generated? https://…/photo.jpg”

Metadata

Remove EXIF, GPS and hidden characters without re-encoding

Ask whether a photo reveals where it was taken, and get a clean copy without the GPS position, camera details, XMP or AI-generator tags. Images and videos are rewritten without their metadata, never decoded or re-encoded, so there is no quality loss.

For text, it finds and removes zero-width characters, direction overrides and the Unicode tag characters that can hide instructions, and decodes what they were hiding. It removes metadata only: watermarks inside the pixels are not touched.

Tools: inspect_metadata, remove_metadata. Try the metadata remover free on the web.

Ask, for example

  • “Does this photo reveal where it was taken? Remove the GPS data if it does.”
  • “Check this text for hidden characters and give me a clean copy.”

Tools and pricing

Free tries, then pay per call

Every account gets 3 free tries on each tool family, shared with the API. Then each call is charged to your prepaid balance. Prices below are read live from the API.

ToolWhat it doesPrice
sign_imageMark an AI-generated imageAdds a signed C2PA manifest, an invisible watermark and a registered fingerprint, and returns a link to the signed PNG.€0.01 / image
verify_imageCheck an image for provenanceLooks for a C2PA manifest, an OpenCorpo watermark and a fingerprint match, and says whether it found all, some or none of the three.€0.002 / image
analyze_videoSummarize a videoFrom a public YouTube link or a video file: summary, timestamped scene description, transcript and a text-to-video prompt.€0.05 / minute
get_video_jobGet a video resultReturns a running or finished analysis. Long transcripts come in pages, so they fit in the conversation.Free
inspect_metadataFind hidden metadataLists EXIF and GPS data, XMP, C2PA and AI-generator tags in a photo or video, and invisible characters in text.€0.003 / image · €0.02 / video · €0.001 / text
remove_metadataRemove hidden metadataReturns a clean copy without re-encoding a pixel or a frame, or the text without hidden characters.€0.003 / image · €0.02 / video · €0.001 / text
get_accountCheck your balanceYour prepaid balance and the free tries you have left on each tool.Free
get_pricingSee the pricesThe price of every operation, read live from the API.Free

Also free: cancel_video_job and create_upload. Video is charged per started minute, once per video whatever outputs are asked for; the quality model costs ×3 and high detail ×2. Failed calls are never charged. All prices.

Remote, on your account

Remote vs local MCP: no install, no blocked transcript scripts

A local MCP server runs on your computer and only in the app that started it. OpenCorpo is a hosted URL: it works in Claude on the web and on your phone, in ChatGPT and in your own agents, with the same billing as the API.

Nothing to install

No npx, no Python, no proxy. YouTube links are fetched by Google's model, not by a script that YouTube blocks on server networks.

Built for agents

Every tool has typed inputs and outputs, and errors say what to do: for example the amount needed and a top-up link when the balance is low.

Free tries, then pay per call

No subscription. Each call is charged at the published prices, and nothing is charged when a call fails.

Confirmed before it costs

Paid tools are declared as actions, so apps that confirm actions ask before each one. Checking your balance and the prices is free.

No key in your chat

Claude and ChatGPT connect with a sign-in and a consent screen on app.opencorpo.com. Developer tools use your API key.

Disconnect in one click

Every connected app is listed in your dashboard. Disconnecting stops it at once.

Never a surprise

Out of balance, a call is refused with the amount needed, never run on credit.

Files come back as links

Signed images and cleaned files are returned as private links that expire after an hour.

FAQ

Questions about the MCP server

Can Claude watch YouTube videos?

Not on its own, but it can with an MCP server. Add OpenCorpo as a connector and give Claude a public YouTube link: the video is analysed by a model that watches the frames and listens to the audio, and Claude gets back a summary, a timestamped scene description, the transcript and a text-to-video prompt.

Can ChatGPT summarize a YouTube video?

Yes, with OpenCorpo added as a custom MCP server. Paste the link and ask for a summary, the key moments with timestamps or the full transcript. You can also attach a video file in the chat.

What is the MCP server URL?

https://api.opencorpo.com/mcp. It is one remote server for all three tools (video, image marking and metadata) over Streamable HTTP. Claude and ChatGPT connect with an OpenCorpo sign-in; developer tools send an API key.

Does it work on YouTube videos without captions?

Yes. It does not read YouTube's captions: the model watches the video and listens to its audio, so it transcribes speech and describes what is shown even when a video has no subtitles. The video must be public.

How do I add a custom connector in Claude?

In Claude (web, desktop or mobile), open Customize, then Connectors, choose Add custom connector and paste https://api.opencorpo.com/mcp. Claude uses OAuth: you sign in to OpenCorpo and approve, with no token to paste. Claude's free plan allows one custom connector. In Claude Code or the Claude API you can use an API key as a bearer token instead.

How do I add an MCP server to ChatGPT?

In ChatGPT's settings, add a custom MCP server with https://api.opencorpo.com/mcp and OAuth, then sign in to OpenCorpo and approve. Custom MCP servers are available on ChatGPT plans that allow them, and the menu names change from time to time.

Is there a free YouTube transcript MCP server?

OpenCorpo gives every account 3 free video analyses (each up to the first 5 minutes), with the summary, scene description and prompt as well as the transcript. After that it is €0.05 per started minute, once per video, from a prepaid balance with no subscription.

What is the difference between a remote and a local MCP server?

A local server runs on your computer (usually started with npx or uvx) and only works in the app that started it. A remote server is a URL: nothing to install, it works in Claude on the web and on your phone, and it does not depend on your network. Transcript scripts run from servers are often blocked by YouTube; OpenCorpo hands the link to Google's model, which fetches the video itself.

Can my agent check an image for C2PA Content Credentials?

Yes. verify_image reports whether an image carries a C2PA manifest, an OpenCorpo invisible watermark and a fingerprint registered by any OpenCorpo customer, and says whether it found all, some or none of the three. That is a technical check, not a legal assessment. Finding no marks does not prove an image was made by a person: metadata is often stripped when images are shared.

Does removing metadata remove AI watermarks?

No. remove_metadata deletes EXIF, GPS, XMP, C2PA and container tags and invisible characters in text, without touching pixels or audio, so watermarks inside the image itself, such as SynthID or OpenCorpo's own, stay where they are. Do not use it to remove AI disclosures you are required to keep.

Is it safe to connect my account?

Chat apps never see a password or an API key: you approve on app.opencorpo.com what the app may do, and every connection is listed under API keys, Connected apps, where you can disconnect it at once. Paid tools are declared as actions, so apps that ask before acting confirm each one.

Where is my content processed?

OpenCorpo runs in Paris on AWS eu-west-3. Videos are analysed by Google's Gemini API on a paid plan, so Google does not use them to improve its models, but that processing is not guaranteed to happen in the EU. Images, text and metadata are processed by OpenCorpo itself.

Connect OpenCorpo to your AI

Create an account and use the free tries from Claude, ChatGPT or your own agent. No card needed until you top up.

Get your free API key →Read the MCP docs