Connect the MCP server to Claude, ChatGPT or Cursor

TranscriptFetch ships an MCP server, so an assistant can fetch a transcript in the middle of a conversation instead of you copying text between windows.

The endpoint

There are two ways to authenticate, depending on the client: OAuth (Claude.ai and ChatGPT open our sign-in page and you approve access, no key involved) or an API key sent as a bearer header (Claude Code, Claude Desktop, Cursor, VS Code). Keys are created in your dashboard and start with tf_live_.

Claude.ai (web)

The fastest path. Requires a paid Claude plan (Pro, Max, Team or Enterprise), which is where custom connectors are available.

1. Open this prefilled link — Claude's Add custom connector dialog opens with the name and server URL already filled in. (Manual path: Settings → Connectors → Add custom connector, name it TranscriptFetch, server URL https://transcriptfetch.com/mcp.)
2. Click Connect. You are taken to TranscriptFetch's sign-in page; sign in (or create an account) and approve access.
3. Done. Paste a video URL and ask for the transcript.

ChatGPT

TranscriptFetch is listed in ChatGPT's plugin directory.

1. Open TranscriptFetch in ChatGPT and click Connect.
2. Sign in to TranscriptFetch (or create an account) and approve access. No API key is needed.
3. Start a new chat and paste a video URL.

If your workspace has not enabled the plugin, add it by hand: in ChatGPT open Plugins → Add → Create custom MCP server, name it TranscriptFetch, server URL https://transcriptfetch.com/mcp, Authentication OAuth.

Claude Code

claude mcp add --transport http transcriptfetch https://transcriptfetch.com/mcp \
  --header "Authorization: Bearer tf_live_YOUR_KEY"

Claude Desktop

Bridge the HTTP server with mcp-remote in claude_desktop_config.json:

{
  "mcpServers": {
    "transcriptfetch": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://transcriptfetch.com/mcp",
        "--header",
        "Authorization: Bearer tf_live_YOUR_KEY"
      ]
    }
  }
}

Cursor / VS Code

Add to your MCP config (Cursor: ~/.cursor/mcp.json, VS Code: .vscode/mcp.json):

{
  "mcpServers": {
    "transcriptfetch": {
      "url": "https://transcriptfetch.com/mcp",
      "headers": {
        "Authorization": "Bearer tf_live_YOUR_KEY"
      }
    }
  }
}

The tools

• get_transcript - one video or media file
• search_videos - find videos by keyword on YouTube, TikTok or Instagram; on YouTube, filter by upload date, length or captions, or sort by views
• search_channels - find YouTube channels or playlists by keyword
• list_channel_videos - a YouTube channel's videos, Shorts or live streams (sorted or searched), or a TikTok or Instagram profile's posts
• list_channel_playlists - a YouTube channel's playlists or podcasts
• list_playlist_videos - the videos in a YouTube or TikTok playlist
• get_credits - your remaining balance, free to call
• create_monitor - watch a YouTube channel or a TikTok or Instagram profile for new videos (a monitor)
• list_monitors and get_new_videos - your monitors, and the new videos they found
• check_monitor_now - run a monitor's check now
• delete_monitor - delete a monitor

The monitor tools change your account, so the assistant should ask you before it creates or deletes one. ChatGPT does not offer them yet.

Any client that speaks MCP works with the same URL; nothing in the server is specific to a particular client.

Troubleshooting

• "Add custom connector" is missing in Claude.ai. Custom connectors require a paid Claude plan; the free plan does not show the option.
• The connect step fails or loops on Claude.ai / ChatGPT. Complete the sign-in and the approve screen in the window that opens. If it still fails, remove the connector and add it again from scratch.
• unauthorized or 401 errors in Claude Code, Cursor or VS Code. The bearer header is missing or the key is wrong. Check the key in your dashboard; it must start with tf_live_ and the header must read exactly Authorization: Bearer tf_live_....
• The tools do not appear after editing a config file. Cursor, VS Code and Claude Desktop read their MCP config at startup; restart the app after saving.
• Verify the connection for free. Ask the assistant to call get_credits; it costs nothing and confirms the server, the auth and your account are all wired.

Billing

Identical to the API: one credit per delivered caption transcript or non-empty listing page. AI Fallback Transcription is billed by audio duration, only on delivery. A monitor costs 1 credit to create and 1 credit for every check, whether or not it finds anything, until you delete it. Failed requests, get_credits, list_monitors, get_new_videos and delete_monitor cost nothing.

The MCP server uses the same key and the same credits as the REST API. If you have not created an account yet, create one first.

Full detail, including copies of every config: the MCP documentation.