Model Context Protocol
Claude MCP server for video & podcast transcripts
Works with Claude Desktop, Claude Code, Cursor, Codex, and any MCP client. Five tools that turn YouTube, TikTok and Instagram into clean, timestamped text your assistant can read and cite.
claude mcp add --transport http transcriptfetch https://transcriptfetch.com/mcp \ --header "Authorization: Bearer tf_live_YOUR_KEY"
Five tools, one server
The server advertises exactly what it can do, so agents plan multi step work on their own: search for videos, list a channel, then pull each transcript.
Full transcript for one video: YouTube ID or URL, TikTok, Instagram, or a direct media file URL. Plain text or timestamped segments.
Search YouTube by keyword. Returns titles, IDs and URLs your agent can feed straight back into get_transcript.
Recent videos for a channel handle (@name), channel ID, or URL.
Every video in a playlist, by ID or URL.
Check the remaining credit balance. Free, so agents can budget before they fetch.
Connect it to your client
One server URL, https://transcriptfetch.com/mcp, over streamable HTTP. Claude.ai signs in with OAuth; everything else sends an API key as a bearer header.
Claude, one-click connector
Settings → Connectors → Add custom connector. Name it TranscriptFetch, set the URL below, click Connect and sign in. No API key: OAuth handles it. Needs a paid Claude plan, where custom connectors live.
https://transcriptfetch.com/mcp
Claude Code MCP server
One command in a terminal, works on any Claude plan:
claude mcp add --transport http transcriptfetch https://transcriptfetch.com/mcp \ --header "Authorization: Bearer tf_live_YOUR_KEY"
Claude Desktop
Desktop speaks stdio, so 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 MCP server
Add the entry to ~/.cursor/mcp.json (VS Code: .vscode/mcp.json), restart, and the tools appear in Agent mode:
{
"mcpServers": {
"transcriptfetch": {
"url": "https://transcriptfetch.com/mcp",
"headers": {
"Authorization": "Bearer tf_live_YOUR_KEY"
}
}
}
}Codex MCP server
Codex configures MCP servers over stdio in ~/.codex/config.toml, so it uses the same mcp-remote bridge as Claude Desktop:
[mcp_servers.transcriptfetch] command = "npx" args = [ "-y", "mcp-remote", "https://transcriptfetch.com/mcp", "--header", "Authorization: Bearer tf_live_YOUR_KEY" ]
Every platform, one MCP server
Most MCP servers wrap one site. This one covers the places spoken content actually lives, with automatic speech-to-text when a video has no caption track.
YouTube MCP
Any video ID or URL to clean text. Channels, playlists and keyword search resolve to video lists, so an agent can sweep a whole channel unattended.
YouTube Transcript API →Instagram MCP
Reel URL in, spoken words out. Instagram has no public transcript endpoint, so the server transcribes the audio automatically.
Instagram Transcript API →TikTok MCP
Most TikToks carry no caption track at all. The server falls through to speech-to-text, so the transcript comes back either way.
TikTok Transcript API →Podcasts & audio files
Point get_transcript at a direct episode audio URL and it transcribes the file. For Spotify and Apple links, the REST API resolves them for you.
Podcast Transcript API →Listed where MCP servers live
The server is published in the official MCP Registry (as com.transcriptfetch), and indexed with an official badge on the major directories:
Where it fits among the best MCP servers
MCP servers compose. A useful agent stack pairs a filesystem or GitHub server for code, a browser or fetch server for the live web, and TranscriptFetch for spoken content: it is the layer that makes video and audio legible. If your agent needs to know what was said in a video, no generic browser server gets there, because the words are in the audio track, not the page.
You: Summarize the argument in this video and pull the three
strongest quotes: youtube.com/watch?v=...
Claude: [calls get_transcript]
The speaker argues that... "direct quote one" (2:41)...Frequently asked questions
What is an MCP server?
MCP (Model Context Protocol) is the open standard AI assistants use to call external tools. An MCP server exposes capabilities, this one exposes video transcripts, and any MCP client such as Claude, Cursor, or Codex can connect to it and use those tools mid-conversation.
Which platforms does the TranscriptFetch MCP server support?
YouTube (including channel, playlist and keyword search resolution), TikTok, Instagram, and direct media file URLs such as podcast episode audio. Videos without captions are transcribed automatically, so you get a result whether or not captions exist.
Does it work with Claude's free plan?
The one-click claude.ai connector requires a paid Claude plan (Pro, Max, Team, or Enterprise), because that is where custom connectors live. Claude Code and Claude Desktop work on any plan using a TranscriptFetch API key.
What does it cost?
TranscriptFetch gives every account 100 free credits a month. One credit per successful fetch, and failed or empty fetches are never billed. Paid plans start at $5 a month for 500 credits.
Do I need an API key?
Not for claude.ai: the connector authenticates over OAuth, you just sign in and approve. Claude Code, Claude Desktop, Cursor, VS Code and Codex use an API key sent as a bearer header.
How is this different from the REST API?
Same engine, different door. The REST API is for your code: batch endpoints, callbacks, SDKs. The MCP server is for your assistant: Claude or Cursor calls the tools itself, with no glue code. Spotify and Apple Podcasts link resolution is currently REST-only.
Give your assistant ears
Sign up, grab a key, paste one config block. 100 free credits a month, one credit per successful fetch, failures never billed.