If you’ve ever pasted a YouTube link into Claude and gotten back a polite “I can’t watch videos,” you know the gap. Claude is great at thinking about content, but it can’t fetch a transcript on its own. The Model Context Protocol fixes that. Once you connect an MCP server that knows how to pull transcripts, Claude can summarize, translate, and analyze YouTube videos as easily as it works with text you paste in.
This guide walks through setting up the TranscriptMagic MCP server in Claude. The whole thing takes about a minute and works in Claude Desktop, Claude.ai on the web, Claude Code, and any other MCP-compatible client like ChatGPT or Cursor.
What is MCP and why it matters for video
The Model Context Protocol is a standard for giving AI assistants access to external tools and data. Think of it as a universal port for language models. Each MCP server exposes a small set of tools (fetch a transcript, look up a customer record, run a database query) and any compliant client can plug in and use them.
Before MCP, the workaround for “summarize this YouTube video” was to copy the transcript by hand, paste it into the chat, and then ask your question. With an MCP server connected, Claude does the fetch itself when you give it a URL. No copy-paste. No browser extension. The video becomes part of the conversation the same way a file would.
That’s the practical win. The bigger one is composition. Once a transcript is in the context window, you can chain it with other tools. Pull a transcript, then ask Claude to draft a LinkedIn post from it. Pull ten transcripts and ask for a competitive analysis. Pull a Spanish-language video and ask for an English summary with timestamps preserved. The transcript is just the starting point.
What you need before you start
You need a Claude account (any tier, including free) and a TranscriptMagic account. New accounts get free credits, so you can test the setup without paying anything. Each successful transcript call uses one credit.
You don’t need an API key. The MCP server uses OAuth, so the first time you call a tool from Claude, a browser tab opens and asks you to sign in. After that, your client remembers the connection and the prompt goes away.
Adding the MCP server to Claude Desktop
Open Claude Desktop and go to Settings, then Connectors. Click “Add custom connector” and paste this URL:
https://mcp.transcriptmagic.com/mcpSave the connector. A browser tab opens for the OAuth handshake. Sign in with the same email you use for your TranscriptMagic account, approve the permissions, and you’re done. The tools show up in any chat as soon as you mention something that looks like a video URL.
To check it’s working, paste a YouTube link into a new chat and ask Claude to summarize it. You should see a tool-use indicator while Claude calls the MCP, then the summary appears with the transcript implicitly included.
Adding the MCP server to Claude on the web
The flow on claude.ai is almost identical. Go to claude.ai/settings/connectors, click “Add custom connector,” paste the same MCP URL, save, and complete the OAuth tab when it opens.
Web Claude tends to be the easiest place to test, because there’s no install step and you can switch between projects without restarting anything.
Adding the MCP server to Claude Code
If you use Claude Code from the terminal, run this once:
claude mcp add --transport http TranscriptMagic https://mcp.transcriptmagic.com/mcpThe first call from inside Claude Code triggers the browser-based OAuth. After that, every project on your machine has access to the same tools.
What you can ask Claude once it’s connected
The most common request is “summarize this video,” which works without any extra prompting. Just paste the URL. But the interesting use cases sit one layer above that.
Translation is a natural fit. Drop a video in any of the 39 supported languages and ask Claude for an English transcript with timestamps preserved. Claude pulls the transcript through the MCP, then translates it inline. You get the original timing data plus a translation in one shot, which is hard to do without an LLM in the loop.
Repurposing is the other big one. Asking for “a LinkedIn post from this video” or “a Twitter thread, four to six tweets, no hashtags” both work cleanly because Claude has the full text to work with rather than a 200-word summary. The output reads like someone watched the video and wrote it, because in effect that’s what happened.
For research, ask Claude to compare a few videos. “Pull these three competitor product launch videos and tell me where they overlap on positioning.” The MCP can fetch each one in turn, and Claude does the synthesis.
Other clients that work with the same MCP server
The same URL works in any MCP-compatible client. ChatGPT supports custom connectors in its settings. Cursor has an MCP section under settings where you add a new server with the URL. The setup is the same shape in every client: paste the URL, complete OAuth, use the tools.
You can connect the same MCP server to multiple clients with one TranscriptMagic account. Credits are shared across all of them.
Common questions
Does this work on YouTube Shorts and live replays?
Yes. Any public YouTube URL works, including youtu.be short links, /shorts/ paths, and replays of live streams. The MCP also supports TikTok, Instagram Reels, and Facebook video.
How much does it cost?
Free credits come with every account, and after that it’s pay-as-you-go at one credit per successful call. There’s no monthly fee for the MCP itself. Pricing details are on the credits page.
Can I use this for commercial work?
Yes. The transcripts come from public videos, and there are no restrictions on what you do with the output. Most paying users build internal research workflows or content tools.
What if I want to call the same tools from my own code?
Use the REST API instead of the MCP. Same data, same auth model, just direct HTTP calls. The MCP is the right pick when you want Claude (or another LLM client) to drive the workflow. The API is the right pick when you’re writing the workflow yourself.
Get started
Open Claude, add https://mcp.transcriptmagic.com/mcp as a custom connector, sign in, and try summarizing your first video. The full walkthrough on the MCP page covers every supported client if you get stuck on a step.