vink
Fetches YouTube captions/transcripts, lists available caption languages, retrieves video metadata and transcript chunks, and saves cleaned, timestamped Markdown transcripts or notes.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Why vink?
Most "YouTube to notes" tools summarize. An hour of teaching becomes a page, and the examples, steps and asides that made the video worth watching are gone.
vink does the opposite. It takes the full transcript YouTube already has, turns it into readable, timestamped text, and hands it to your AI assistant one part at a time, so nothing is lost when notes are written.
Typical summarizer | vink | |
A 1-hour lecture becomes | A few bullet points | Complete notes, section by section |
Examples and steps | Mostly dropped | Kept, in order |
Jump back to the video | Rarely | Every paragraph links to its timestamp |
Your notes app | Copy and paste | Markdown with front matter, ready for Obsidian |
Setup | Account and API key |
|
Related MCP server: YouTube Data MCP Server
Features
Quickstart
1. Install (Python 3.11+):
uv tool install vink # or: pipx install vink2. Get a transcript:
vink transcript https://youtu.be/aircAruvnKk/your/folder/output/But what is a neural network Deep learning chapter 1.md3. Open the file. It looks like this:
---
title: "But what is a neural network? | Deep learning chapter 1"
channel: "3Blue1Brown"
language: "en"
captions: "manual"
duration: "18:25"
---
# But what is a neural network? | Deep learning chapter 1
**Channel:** 3Blue1Brown · **Duration:** 18:25 · **Language:** English (creator captions)
## Transcript
**[0:04](https://youtu.be/aircAruvnKk?t=4)** This is a 3. It's sloppily written and rendered
at an extremely low resolution of 28x28 pixels, but your brain has no trouble ...
**[0:49](https://youtu.be/aircAruvnKk?t=49)** But if I told you, hey, sit down and write for me
a program that takes in a grid of 28x28 pixels ...Use it with your AI assistant
vink includes an MCP server. Your assistant reads the transcript part by part and writes full notes, with no API key and nothing extra to pay.
Claude Desktop
Add this to claude_desktop_config.json
(macOS: ~/Library/Application Support/Claude/, Windows: %APPDATA%\Claude\),
then restart Claude:
{
"mcpServers": {
"vink": {
"command": "uvx",
"args": ["vink", "mcp"]
}
}
}Then ask:
Make detailed notes from https://youtu.be/aircAruvnKk in Hinglish and save them.
Or choose + → Add from vink → make_notes.
claude mcp add vink --scope user -- uvx vink mcpAdd to ~/.cursor/mcp.json:
{
"mcpServers": {
"vink": { "command": "uvx", "args": ["vink", "mcp"] }
}
}code --add-mcp '{"name":"vink","command":"uvx","args":["vink","mcp"]}'Most clients accept the standard block:
{
"mcpServers": {
"vink": { "command": "uvx", "args": ["vink", "mcp"] }
}
}If the client cannot find uvx, use its full path (run which uvx).
What your assistant can do
Tool | What it does |
| Lists the caption languages a video has |
| Title, language, duration and number of parts |
| One part of the transcript, as Markdown |
| The whole transcript, for short videos |
| Saves the transcript as a Markdown file |
| Saves the notes your assistant wrote |
Two note styles are included: detailed-notes, complete study notes (the
default), and clean-transcript, an edited transcript that stays close to
word for word. Files are saved to ~/Documents/vink unless you set
VINK_OUTPUT_DIR.
More details: docs/mcp.md.
Command line
vink transcript URL Save a transcript as Markdown
vink languages URL List the caption languages a video has
vink mcp Run the MCP server# Hindi if available, otherwise English
vink transcript https://youtu.be/VIDEO_ID --lang hi,en
# The language spoken in the video, printed to the terminal
vink transcript VIDEO_ID --lang original --stdoutvink transcript URL [--lang en,hi] [--output-dir DIR] [--format markdown] [--stdout]
[--translate | --no-translate] [--cache | --no-cache]
vink languages URL [--cache | --no-cache]
vink mcp [--transport stdio|http] [--host HOST] [--port PORT] [--insecure-no-auth]
vink [-v | --verbose] [--version] COMMANDOption | Meaning |
| Preferred languages, most preferred first. |
| Print instead of saving a file. |
| Never fall back to YouTube's machine translation. |
| Fetch fresh data from YouTube. |
| Show progress logs (place it before the command). |
Exit codes: 0 success, 1 the transcript could not be fetched, 2 invalid
input or options.
Running on a server
YouTube blocks most requests from cloud servers (AWS, GCP, hosted sandboxes). On your own computer everything just works. For servers, pick one of these:
Host at home | Residential proxy | |
Idea | Run vink on a home machine; connect to it over HTTPS | Run vink anywhere; send YouTube traffic through a proxy |
Cost | Free | Paid proxy plan |
Key setting |
|
|
Host at home:
VINK_AUTH_TOKEN='a-long-random-secret' vink mcp --transport http --port 8000Residential proxy:
VINK_PROXY_URL='http://user:pass@proxy.example.com:8080' vink mcpWebshare rotating residential proxies are supported out of the box via
VINK_WEBSHARE_USERNAME and VINK_WEBSHARE_PASSWORD.
The step-by-step guide covers Cloudflare Tunnel, systemd and troubleshooting: docs/hosting.md.
Configuration
Everything works without configuration. To change a default, set an
environment variable or put it in a .env file:
Variable | Default | What it does |
|
| Preferred languages, comma-separated |
|
| Where files are saved |
|
| Length of one part for |
| none | Proxy for requests to YouTube |
| none | Token required by the HTTP server |
All settings: docs/configuration.md.
How it works
YouTube URL
│
├─ pick the best caption track (your languages · creator > auto > translated)
├─ fetch and cache it
├─ clean it into paragraphs (noise removed · timestamps kept)
├─ split it into ~10-minute parts
└─ Markdown file ──or── MCP tools → your AI assistant → detailed notesFetching, rendering and saving each sit behind a small interface, so new formats (PDF, DOCX) and destinations (Notion, Obsidian) can be added without touching the core. See docs/architecture.md.
FAQ
No. The official YouTube Data API only lets you download captions for videos
you own. vink reads the same caption data YouTube's own player uses, through the
open-source
youtube-transcript-api
library.
Not yet. vink needs either creator-uploaded or auto-generated captions. Transcribing the audio of videos without captions is planned.
Yes. The transcript can be in any language the video has, and the
make_notes prompt can write the notes in any language you name, including
Hinglish.
You are running vink on a cloud server, and YouTube blocks data-center IP addresses. See Running on a server.
Contributing
Contributions are welcome: bug reports, new output formats, destinations and note styles. See CONTRIBUTING.md to get started.
License
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Real-time YouTube transcripts, comments, channels & video search MCP. Local npx or hosted OAuth.
Real-time YouTube video transcripts, auto-translation, metadata, search & comments MCP ($1.00/1K).
1Real-time YouTube video transcripts, auto-translation, metadata, search & comments MCP ($1.00/1K).
Your YouTube library, in the AI tools you already use: Claude, ChatGPT, Cursor, any MCP client.
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables users to extract, search, and analyze YouTube video transcripts directly within MCP-compatible clients. It supports advanced features like time-chunked summaries, keyword searching with surrounding context, and batch processing for multiple videos.4MIT
- FlicenseAqualityDmaintenanceEnables extracting YouTube video transcripts, comments, and search results via MCP. Supports both Stdio and StreamableHTTP transports.518 npm2-
- AlicenseNot gradedqualityCmaintenanceEnables MCP-compatible assistants to retrieve caption text from YouTube videos as either plain text or timestamped snippets for reading, summarizing, or searching. Supports local stdio and stateless HTTP deployments for integration with Claude Desktop, Claude Code, Cursor, and similar clients.MIT
- AlicenseAqualityCmaintenanceEnables MCP clients and agents to fetch YouTube transcripts and caption lists from video IDs or links, preferring manual/auto captions by ordered language preference and falling back to local Faster-Whisper audio transcription when captions are unavailable. Results include source, language, full text and timestamped segments, with CLI exports to JSON, text, SRT and WebVTT, all without paid APIs or API keys.3MIT