# VidWords YouTube [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/haljishi/vidwords-mcp  
**GitHub Stars:** 0  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/vidwords-youtube-2

## Description
Search YouTube transcripts and read a video's frames; answers cite clickable timestamps.

## Claude Desktop Quick Installation
No confirmed install details yet, so no config is published. Follow the setup steps in the repository above.

## Documentation

## What VidWords YouTube MCP server does

VidWords YouTube MCP server gives an AI client access to information contained in YouTube videos without requiring the client to process the video directly. Its capabilities cover transcript search, searches over videos already saved to a user’s Library, and inspection of individual video frames. The service can return answers with citations that point to the relevant second in the source video.

Frame access is intended for visual material that transcripts cannot represent well, including slides, charts, demonstrations, and text shown on screen. The repository is a public manifest and configuration reference for the hosted service rather than a locally implemented video-processing backend.

## How it works

The hosted MCP endpoint is `https://vidwords.com/mcp`. MCP-compatible clients connect to that URL and call the exposed tools. The service uses OAuth for interactive clients: the client registers, opens a browser sign-in flow, and stores credentials for later use. A user can create an account during that process, and access begins after approval.

Headless clients and automation can authenticate with a static API token sent as `Authorization: Basic YOUR_API_TOKEN`. The repository also includes a small stdio proxy for clients that cannot add custom HTTP headers. That proxy forwards calls to the hosted endpoint and can be run through `npx` from the GitHub repository or in Docker.

## Setup and configuration

For clients with HTTP MCP support, configure the server URL as `https://vidwords.com/mcp` and complete the client’s OAuth flow. The README documents setup paths for Claude, Claude Desktop, ChatGPT, Claude Code, Cursor, VS Code, and Codex CLI. Claude Desktop’s configuration file requires the `mcp-remote` bridge because its direct configuration supports local stdio servers rather than a remote URL.

For token-based access, create and verify an account, generate an API key from the VidWords API and MCP page, and send it with the Basic authorization scheme. Keep that value out of committed configuration files. The stdio proxy reads `VIDWORDS_API_TOKEN`; `VIDWORDS_MCP_URL` can override the default upstream endpoint.

## Tools and capabilities

VidWords YouTube MCP server provides twelve tools. The supplied material identifies these tool groups:

- Search video transcripts for relevant spoken content.
- Search videos already stored in the user’s Library.
- Read selected video frames for visual context.
- Answer questions with verified, clickable timestamp citations.
- Transcribe audio and analyze visual frames through the hosted service.

The service separates usage into Cloud Requests and AI Units. Fetching a video transcript uses a Cloud Request. Frame reading and audio transcription consume AI Units; a newly analyzed frame also involves fetching the video transcript and AI usage based on video duration. Re-reading a video already in the account’s Library is described as free, including another time range.

## Limitations and notes

VidWords YouTube MCP server is remote-only in its normal form: there is no production server package to install or self-host in this repository. The included proxy is an adapter, not a replacement for the hosted endpoint. Its GitHub-based `npx` invocation should not be replaced with an assumed npm package name.

Most clients can use OAuth without manually copying a token, but clients that lack OAuth or custom-header support need the static-token route or the stdio proxy. Basic authentication is required for the token route; using a bearer-token configuration sends the wrong authorization scheme.

The free plan is described as including 25 Cloud Requests and 200 AI Units per month. Usage therefore depends on which operations an agent performs, especially transcript fetching, frame analysis, and audio transcription.

_Full upstream README: https://allmcps.com/mcp/vidwords-youtube-2/readme_