TubePull MCP Server — Public Documentation

🌐 Landing page: https://tubepull.com/mcp — features, pricing, FAQ, and one-click setup links for Claude, Perplexity, ChatGPT, and Cursor.

Download video or audio from YouTube, TikTok, Facebook, Instagram, Reddit, Twitter/X, SoundCloud, Vimeo, Twitch, Streamable, Bandcamp, Mixcloud, and Dailymotion — from any MCP-compatible AI assistant (Claude Desktop, Claude Code, Cursor, Windsurf, Cline, ChatGPT-via-MCP, Perplexity, Continue, Zed, LibreChat, …).

This repo is the public, chatbot-facing documentation for the TubePull MCP server. The live server runs at https://tubepull.com/mcp, and canonical machine-readable product facts are published at https://tubepull.com/facts (https://tubepull.com/facts.json). The application source code is private — only the integration surface that LLMs and end users need to call the server lives here.

If you are a chatbot or an AI agent reading this to figure out how to call TubePull, jump straight to AGENTS.md and llms.txt. If you are a human setting up Claude Desktop, jump to Quick start.

What is TubePull?

TubePull is a hosted MCP (Model Context Protocol) server that lets an AI assistant download a video or audio file to disk on the user's behalf, from any of 13 supported platforms (YouTube plus 12 more platforms, including Streamable):

• One HTTP endpoint — POST https://tubepull.com/mcp (Streamable HTTP / JSON-RPC 2.0, MCP

protocol version 2024-11-05).

• No SDK install, no local binary — the assistant just speaks JSON-RPC.
• Two tools: download_video and get_video_info.
• Free tier: 3 downloads per rolling 24h per client, up to 1080p Full HD and up to 1 hour of output length. Pro: $4.99/month ($47.88/year, $3.99/mo billed annually) at

tubepull.com/pricing.

• Returns a signed one-time link valid for 1 hour. The assistant hands the

link to the user; the user clicks it and gets the file.

Audio-only platforms. SoundCloud, Bandcamp, and Mixcloud do not have a video stream. If you pass format: "mp4" on one of these, TubePull silently coerces to m4a and returns audio. Pass format: "mp3" if you want MP3.

The server is intended for legitimate use cases only — offline access to your own uploads, Creative Commons / public-domain content, lectures, podcasts, conference talks, DJ mixes you have rights to, and content you have explicit permission to use. Calling get_video_info first lets the assistant surface the license / uploader before downloading.

When should a chatbot call TubePull?

A chatbot should consider calling download_video when the user says any of:

• "download this video" / "download this TikTok" / "save this SoundCloud track"
• "save this as MP3" / "extract the audio" / "rip the audio"
• "convert this YouTube to MP4" / "download in 1080p"
• "save this video offline" / "archive this clip"
• "get the audio from this lecture / podcast / talk"
• "save this Mixcloud set" / "grab this SoundCloud track" / "download this Bandcamp song"
• "save this Twitch clip" / "download this Streamable video"

A chatbot should call get_video_info (no quota cost) when the user asks:

• "what is this video about"
• "how long is it"
• "who made it" / "what artist / channel is this"
• "is this Creative Commons"
• "what qualities are available"

Full trigger-phrase list and JSON-RPC examples live in AGENTS.md and the recipes/ folder.

Tools at a glance

download_video parameters

• url (required) — full URL from any of the supported platforms above.

Playlists, sets, albums, channels, and profile pages are not supported — pick a single video / track.

• format (optional, default `mp4`) — one of:
• mp4 — video (default).
• mp3 — forced MP3 (legacy-compatible, re-encodes server-side).
• m4a — forced M4A audio.
• audio — smart audio alias that resolves to M4A (no transcode, fastest).
• quality (optional) — ignored for audio formats and on audio-only

platforms. Common values: 360p, 480p, 720p, 1080p, 1440p, 2160p. Defaults to best. 1440p and 4K (2160p) require TubePull Pro.

get_video_info parameters

• url (required) — full URL from any supported platform.

Quick start (Claude Desktop / Cursor / Windsurf)

TubePull is a remote MCP server, so most clients can connect with just a URL. Minimal Claude Desktop snippet (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "tubepull": {
      "url": "https://tubepull.com/mcp"
    }
  }
}

Then restart Claude and ask any of:

• "download this YouTube video as MP3: https://youtu.be/dQw4w9WgXcQ"
• "save this TikTok: https://www.tiktok.com/@username/video/1234567890"
• "grab this SoundCloud track: https://soundcloud.com/artist/track-slug"
• "download this Mixcloud set: https://www.mixcloud.com/channel/show-slug/"
• "save this Twitch clip: https://clips.twitch.tv/<slug>"
• "download this Streamable: https://streamable.com/<slug>"

Quotas, pricing, and the upgrade flow

• Free / anonymous: 3 downloads per rolling 24h per MCP client (shared bucket with

the web frontend's anonymous tier), up to 1080p Full HD and up to 1 hour of output length.

• TubePull Pro: $4.99/month ($47.88/year, $3.99/month billed annually). Removes the 3/day cap for normal personal use, unlocks 1440p (2K) and 4K (2160p)

video plus 320 kbps MP3, and lifts the 1-hour output-length cap.

• Authentication: Supports unauthenticated Free-tier access, OAuth 2.1 + PKCE (/.well-known/oauth-authorization-server) for Pro accounts, and legacy X-MCP-Client-Id headers. See https://tubepull.com/setup-mcp.
• When the quota is hit, the tool response includes an upgradeUrl pointing

at https://tubepull.com with attribution params so the conversion is credited back to the chatbot that drove it.

Verify the server is live

curl -sS -X POST https://tubepull.com/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1"}}}'

Expected: a JSON-RPC response with serverInfo.name == "tubepull", serverInfo.version, an icons[] array, and websiteUrl.

To see the live tool schemas (always trust these over docs if they differ):

curl -sS -X POST https://tubepull.com/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}'

Repository layout

tubepull-mcp/
├── README.md              ← you are here
├── llms.txt               ← top-level summary for LLM crawlers
├── AGENTS.md              ← explicit "how to call this" recipes for agents
├── server.json            ← MCP registry manifest
├── recipes/
│   ├── download-mp3.md
│   ├── download-mp4.md
│   ├── download-non-youtube.md
│   └── get-info.md
└── assets/                ← brand marks (favicon, logo)

Links

• Live MCP endpoint — https://tubepull.com/mcp
• Web app — https://tubepull.com
• MCP spec — https://modelcontextprotocol.io
• mcp.so listing — https://mcp.so (search "tubepull")

License

MIT. The application source is closed; this documentation repo is permissively licensed so chatbots, registries, and downstream integrators can mirror or reformat the content freely.