FastClip Logo
fastclips.app

AI AGENT RECIPE

Read a YouTube transcript with MCP, then cut the moment

Most YouTube transcript MCP servers stop at the text. FastClip's server gives your agent the transcript with times in seconds, and then cuts the range the agent chose into a finished clip.

Paste this into your agent

Use FastClip. Import (paste the YouTube link) and read its whole transcript, page by page. Find the part where (describe the idea in your own words), even if the speaker uses other words. Tell me the start and end you chose and why, then make one clip of that range with captions.

Replace the parts in brackets. Your agent must be connected to FastClip first (next section).

Connect your agent once

In Claude Code, Codex, Cursor or VS Code, use a token. Sign in to FastClip and open Settings, then Agents. Create a token and copy it. FastClip shows it only one time. Then add FastClip's MCP server to your agent with the command or install link on the AI agents page. In an app that adds a custom MCP server by its address, add https://api.fastclip.it/mcp and sign in to FastClip when the app asks. No token is needed. A token, or an app you connect, can import videos, read transcripts and make clips. It cannot post, change your settings or delete anything.

What the agent calls, in order

These are FastClip's real MCP tools and arguments. The values are examples. Times are in seconds.

  1. 1. Import the video

    The agent adds the video to your account. FastClip fetches the transcript from the video's YouTube captions.

    import_video { "url": "https://www.youtube.com/watch?v=YOUR_VIDEO" }

  2. 2. Read the first part

    The agent reads from the start. One call returns up to 10 minutes or 12,000 characters, as lines with start and end times.

    get_transcript { "video_id": "VIDEO_ID", "start": 0 }

  3. 3. Read on

    The result says where the next part starts. The agent continues from there until it has read the part it needs.

    get_transcript { "video_id": "VIDEO_ID", "start": 598.2 }

  4. 4. Cut the clip

    FastClip's own search finds words, not ideas. Here the agent reads the text itself, so it can find a moment by meaning. It then cuts the range it chose.

    create_clip { "video_id": "VIDEO_ID", "start": 1022.5, "end": 1101, "format": "original", "captions": true }

  5. 5. Get the link

    The agent waits for the render and gives you the download link.

    get_clip { "clip_id": "CLIP_ID" }

What you get

The agent's reason for the range, and one captioned MP4 of it in your FastClip library.

Where it works

Claude Code
Tested. We made clips from Claude Code with FastClip's server.
Codex
Tested. We made clips from Codex with FastClip's server.
Cursor
The setup follows Cursor's documented format for a remote server. We have not made a clip from Cursor yet.
VS Code
VS Code accepts the setup. We have not made a clip from VS Code yet.
Claude (web and desktop app)
Add FastClip as a custom connector with the server address https://api.fastclip.it/mcp and sign in to FastClip. We tested the sign-in with a generic client, not yet from the Claude app. FastClip is not in the Claude connector directory yet.
ChatGPT
In developer mode, add the server address https://api.fastclip.it/mcp and sign in to FastClip. We tested the sign-in with a generic client, not yet from ChatGPT. FastClip is not in the ChatGPT directory yet.
Grok
Add a custom connector with the server address https://api.fastclip.it/mcp and sign in to FastClip. We tested the sign-in with a generic client, not yet from Grok. FastClip is not in the Grok directory yet.

Limits

  • The video is imported first, so it counts toward your plan's imports, even if you only wanted the text.
  • The transcript is the YouTube captions. A video without captions has no transcript here. If the result says "fetching", the agent asks again after about 15 seconds.
  • A long video takes many calls to read. Each call adds text to your agent's context, which can cost you more with your agent.
  • You choose the moment. FastClip does not cut a batch of clips on its own and gives no virality scores. The agent makes the clips you ask for.
  • Clip length: 1 second to 5 minutes. Original keeps the frame of the video. Vertical is 9:16 at 1080 × 1920.
  • Speed limits: 6 new videos and 20 clips in 10 minutes for your account. The agent gets a clear error with the time to wait.
  • Your plan: the clips and imports an agent makes count toward your plan. The free plan has 100 finished clips and 50 YouTube or upload imports in total, and up to 30 finished clips a day.
  • If YouTube refuses the download: YouTube refuses many downloads from FastClip's servers. Then the import is queued and FastClip tries again by itself for about a day. You can also open the video in FastClip and upload the file. Waiting clips then continue from the file.

Questions

Is this a YouTube transcript MCP server?

It reads transcripts, but it is a clipping server. Use it when you want the clip, not only the text. Each video you read is imported into your FastClip account.

Can it search the transcript?

Yes, by keyword: it returns up to 8 passages that contain the words, with one typo allowed per word. To find an idea by meaning, let the agent read the transcript, as in this recipe.

Which languages work?

The language of the video's YouTube captions. The result says the language.

More recipes

All recipes and the setup

Try it with your own video

Free, no credit card, no watermark. Connect your agent with a token from Settings or with a sign-in, and paste the prompt.

Start clipping for free