Skip to main content
Glama
mcp-youtube-transcript

mcp-youtube-transcript

by gridivit

mcp-youtube-transcript

An MCP server that takes a YouTube video URL and returns the transcript text to your agent.

Tools

  • list_transcript_languages(url) — which subtitle tracks the video has

  • get_transcript(url, language) — the transcript text in the given language

language is required. If the requested language is unavailable, English is used as a fallback; if neither exists, the tool returns the list of languages the video does have so the agent can retry.

URLs are accepted in every common form: watch?v=, youtu.be/, /shorts/, /embed/, /live/, or a bare 11-character video ID.

Related MCP server: YouTube Insights MCP Server

Installation

No Python needed. The only requirement is uv — a single binary that installs without administrator rights and downloads a suitable Python for you.

Windows (PowerShell):

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

macOS / Linux:

curl -LsSf https://astral.sh/uv/install.sh | sh

Then add the server to Claude Code with a single command:

claude mcp add youtube-transcript -- uvx --from git+https://github.com/gridivit/mcp-youtube-transcript mcp-youtube-transcript

For Claude Desktop, add this to claude_desktop_config.json (%APPDATA%\Claude\ on Windows, ~/Library/Application Support/Claude/ on macOS):

{
  "mcpServers": {
    "youtube-transcript": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/gridivit/mcp-youtube-transcript",
        "mcp-youtube-transcript"
      ]
    }
  }
}

For Antigravity, use the same JSON in ~/.gemini/config/mcp_config.json (global) or .agents/mcp_config.json (per project).

The first launch takes about a minute while uv downloads Python and the dependencies. Every launch after that is instant.

Windows notes

  • Escape backslashes in JSON paths: C:\\Users\\..., or just use forward slashes. A single backslash silently breaks config parsing.

  • If the client cannot find uv, restart the application completely so it picks up the updated PATH. If that does not help, use the full path to the binary: C:\\Users\\<name>\\.local\\bin\\uvx.exe.

Updating

uvx caches the cloned repository and will not pick up new commits on its own. To pull the latest version:

uvx --refresh --from git+https://github.com/gridivit/mcp-youtube-transcript mcp-youtube-transcript

Development

Do not install your own checkout through uvx — it pulls the code from git, so you would not see local changes. Point the client at the working copy instead:

claude mcp add youtube-transcript-dev -- uv --directory <path-to-clone> run mcp-youtube-transcript

Run the test client to inspect the protocol directly:

npx @modelcontextprotocol/inspector uv run mcp-youtube-transcript

Available Tools

2 tools
get_transcriptA

Get the full transcript text of a YouTube video.

Returns the whole transcript as plain text without timestamps. Long videos produce long text: a two-hour video is roughly 20-30 thousand words.

If the requested language is unavailable, English is used as a fallback. If neither exists, this fails and lists the languages the video does have - pick one of them and call again.

Args: url: YouTube video URL (watch, youtu.be, shorts or embed form), or the bare 11-character video id. language: Language code of the wanted transcript, e.g. "en", "ru", "de".

Returns: The transcript as a single plain-text string.

Inputs: url (required), language (required). Returns: result

list_transcript_languagesA

List the subtitle tracks available for a YouTube video.

Call this when you do not know which languages a video has, or after get_transcript reported that the requested language is unavailable.

Args: url: YouTube video URL (watch, youtu.be, shorts or embed form), or the bare 11-character video id.

Returns: One line per track: " - (manual|auto-generated)".

Inputs: url (required). Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.7/5.0

    Scored across 2 tools

    Disambiguation5/5

    The two tools have distinct purposes: one lists available subtitle tracks, the other retrieves a transcript in a specified language. There is no overlap in functionality, and the descriptions clarify when to use each.

    Naming Consistency5/5

    Both tool names follow a consistent verb_noun pattern: 'list_transcript_languages' and 'get_transcript'. This creates a predictable and readable convention.

    Tool Count4/5

    Two tools is slightly below the typical 3-15 range, but it is well-scoped for a focused YouTube transcript server. The count feels appropriate rather than thin, covering the core use cases.

    Completeness5/5

    The tool surface covers the essential lifecycle: discovering available languages and fetching the transcript. The fallback to English and error messaging that suggests other languages complement the workflow, leaving no obvious gaps.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers