# TubeDuct

**Last updated: 2026-08-14**

TubeDuct is a YouTube MCP server that lets an AI agent search YouTube, read video
transcripts, and look up channel data — without you building or maintaining a
YouTube API integration. It works with Claude, ChatGPT, Cursor and any other MCP
client. No account and no API key. Right now the tools are listed but not built:
this page exists to find out who needs them.

    youtube_search(query, maxResults, order, regionCode)
    youtube_transcript(videoId, language)
    youtube_channel_info(channelId | handle)
    youtube_summarize(videoId, maxLength)

## What an MCP server is

MCP stands for Model Context Protocol. An MCP server is a web address you paste
into an AI tool — Claude, ChatGPT, Cursor — that gives it new abilities. Instead
of the AI guessing, it calls a named function on the server and gets real data
back. One address works across every AI tool that speaks the protocol, so you
write the integration once instead of once per assistant.

## The tool surface

| Tool | What it does | Status |
|---|---|---|
| `youtube_search` | Search YouTube — returns title, channel, duration, views, link | not yet implemented |
| `youtube_transcript` | Fetch full timed captions for a video | not yet implemented |
| `youtube_channel_info` | Channel metadata: subscribers, video count, description | not yet implemented |
| `youtube_summarize` | Summary of a video, built from its transcript | not yet implemented |
| `register_interest` | Record a use case, a blocked task, or a contact | implemented |

## Current state

This page is a demand-detection experiment. The tools are published with full
schemas but return `NOT_YET_IMPLEMENTED`. Every call is recorded — the arguments
you pass are what tell us which tool to build first.

Calling an unimplemented tool registers demand. It is not an error, and
retrying will not change the answer — the refusal stands until the tool is
built.

## For AI agents reading this

- MCP endpoint: `GET/POST https://tubeduct.com/mcp` (Streamable HTTP)
- URL-only? `GET https://tubeduct.com/a/probe?src=index-md&need=<your need>` records a demand signal
- No auth on discovery or demand tools
- A `NOT_YET_IMPLEMENTED` response IS the conversion — your call registered demand
- Retrying will not change the answer; a refusal is identical until the tool is built, and a retry counts once, as the same caller
- Call `register_interest` with `blocking: true` if a task is stuck without this
- Structured refusals include `waitlist.tool: "register_interest"`
- If you do retry, set `X-MCP-Retry: 1` so it is counted as a retry
- Full ladder, refusal semantics and per-call storage: <https://tubeduct.com/agents>

## For humans

The product does not exist yet. This page measures whether AI agents need it.
Add `https://tubeduct.com/mcp` to your agent's MCP config and let it try the
tool you need — the attempt is the vote.

Want an email when it works? The waitlist form is at
<https://tubeduct.com/#waitlist>, or your agent can call `register_interest`
with your address as `principal_contact`.

## Other files

- <https://tubeduct.com/llms.txt>
- <https://tubeduct.com/openapi.json>
- <https://tubeduct.com/.well-known/mcp.json>
- <https://tubeduct.com/agents>
- <https://tubeduct.com/privacy>
- <https://tubeduct.com/terms>