The reason to prefer this over a local transcript scraper is that research questions usually need discovery before extraction — find the videos, then read them — and both halves live behind one endpoint with one key. `get_channel_latest_videos` is free because it reads RSS, which makes daily upload monitoring viable as a habit rather than a budget line.
A hosted YouTube toolkit at transcriptapi.com. Most YouTube servers fetch one transcript; this one adds search, in-channel search, full channel listing, playlist extraction and RSS-based upload tracking, all remote so there is nothing to install locally.
- `get_youtube_transcript` returns a transcript as markdown with metadata or as JSON, with `include_timestamp` and `send_metadata` both defaulting to true
- `search_youtube` searches videos or channels with a continuation token for paging, about 20 results per page
- `get_channel_latest_videos` returns roughly the 15 most recent uploads for a handle, URL or channel id — free, no credits charged
- `search_channel_videos` searches inside one channel, about 30 results per page
- `list_channel_videos` lists a channel's whole library, about 100 per page
- `list_playlist_videos` returns every video in a playlist, about 100 per page
- One credit equals one successful request; failed and rate-limited calls do not consume credits
A TranscriptAPI account — the free tier gives 100 credits at 60 requests per minute. Authenticate either with an API key beginning `sk_`, sent as a bearer token, or over OAuth 2.1: Claude Desktop, Claude Web and ChatGPT can register dynamically, so adding the URL is the whole setup. The endpoint is https://transcriptapi.com/mcp over Streamable HTTP; Claude Code registers it with `claude mcp add --transport http`. The repository is also an Agent Plugins 1.0.0 package containing no credentials, bundling the MCP server plus a skill that teaches an agent when to use each tool.
One command — claude mcp add --transport http transcript-api https://transcriptapi.com/mcp