Connect YouTube to Claude Desktop
Claude cannot watch a video. One connector lets it read one — and cite the moment it got the answer from.
The short answer
Add to Claude opens Claude with the name and server URL filled in. Review the URL and confirm. If the form does not open, go to Customize → Connectors → + Add → Add custom connector and paste the server URL:
https://vidwords.com/mcp
Under the OAuth client options choose Register automatically — VidWords registers each connector itself, so Claude's default published-identity option will not connect. Then sign in.
On a Team or Enterprise plan an owner adds it first, under Organization settings → Connectors → Add → Custom → Web; each member then connects from Customize → Connectors.
Claude hands you to VidWords to sign in or create a free account. If your address needs confirmation, open the link in your inbox or enter its 6-digit code in the VidWords banner; you can resend the email there. Return to the consent screen and approve access. If the connection flow expired while you checked your email, start it again in Claude. The tools then appear in the composer. No config file or token to copy is needed.
A config file instead
claude_desktop_config.json only runs local servers — programs Claude Desktop starts itself. A remote url entry in it is not supported, and has been reported to wipe the file's whole mcpServers section on startup, so do not paste the endpoint there directly. To declare the server in the file anyway, bridge it with mcp-remote, which runs locally and opens the same sign-in in your browser — still no token to paste:
{
"mcpServers": {
"vidwords": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://vidwords.com/mcp"]
}
}
}
The file lives at ~/Library/Application Support/Claude/claude_desktop_config.json on macOS and %APPDATA%\Claude\claude_desktop_config.json on Windows. It needs Node.js installed, because npx runs the bridge.
A token instead of signing in
Only a machine that cannot open a browser needs this. Pass the key through an environment variable rather than inside args — Claude Desktop on Windows does not escape spaces in args, so Basic YOUR_API_TOKEN written there splits in two:
{
"mcpServers": {
"vidwords": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://vidwords.com/mcp",
"--header", "Authorization:${VIDWORDS_MCP_AUTH}"],
"env": { "VIDWORDS_MCP_AUTH": "Basic YOUR_API_TOKEN" }
}
}
}
Create a free account and create a key on the API & MCP page — its value is shown once, when you create it. Two things to know before the first call:
- Verify your email. Until you click the verification link every call returns
403with{"error":"email_unverified"}— the most common first-call failure on a new account. - The scheme is
Basic, notBearer, and the token goes in as-is. You are not base64-encoding auser:passpair.
Restarting actually matters
Claude Desktop reads this file at launch. Closing the window is not enough on macOS — quit the application properly (⌘Q) and reopen it, or the old configuration stays loaded and you will conclude the config is wrong when it is merely stale.
Once it restarts, the tools appear in the tool menu in the message composer. If the server failed to start, Claude Desktop reports it there rather than silently omitting the tools.
Merging with servers you already have
If mcpServers already exists, add vidwords as another key inside it rather than pasting a second mcpServers block. A second block is still valid JSON, which is what makes it dangerous: the last one silently wins, and every server in the first one is dropped without an error. A trailing comma is different — that makes the file invalid, and an invalid config usually presents as every server disappearing at once. When servers vanish after an edit, check the JSON before you check anything else.
What it is good for
Claude Desktop is where most people do reading and research rather than coding, and video is the format that research workflows handle worst — you cannot skim it, quote it, or search it without watching it.
- "What does this 90-minute interview say about hiring? Give me the timestamps." —
search_transcriptreturns the relevant stretches with clickable deep links, so you can verify each claim in a second rather than trusting a summary. - "Compare what these three videos say about the same product." — a batch through
get_transcript, then ordinary analysis. - "The chart at the end of this talk — what were the actual numbers?" —
analyze_videoreads frames, so on-screen figures that are never spoken aloud are available.ask_videothen answers with citations that are checked against the recorded frames, and says the evidence is insufficient rather than guessing when nothing matches.
The 22 tools
| Tool | What it does | Cost |
|---|---|---|
search_transcript | Find where a video — or a list of up to 25 — discusses something; returns timestamps and deep links. | 1 Cloud Request per new video |
get_transcript | Full text for up to 25 videos, a span between timecodes, or a page at a time with maxChars. | 1 Cloud Request per new video |
translate_transcript | One video’s transcript translated into any language by AI, line for line, every timestamp kept — for a language the video has no captions in, or an audio transcription. estimateOnly quotes the price first; reading it again is free. | 2 AI Units per started minute of video, only when it succeeds |
search_library · list_library | Search and list the transcripts you have already saved. | Free |
list_languages | A video's caption tracks, and which ones transcribe the speech. | Free |
list_channel_videos | Recent uploads for a channel. | 1 Cloud Request · Starter and up |
list_watchlists · watchlist_activity | Radar monitoring — channels you track and what they published. | Free |
create_watchlist · add_watchlist_channel · check_watchlist · preview_watchlist_filters | Set up Radar: create a watchlist, add channels, check them for new uploads now, and test title filters before saving them. | Free |
update_watchlist · set_radar_paused | Rename, pause or filter a watchlist, or pause all of Radar. Turning on automatic analysis spends AI Units at Watch rates, so an assistant should only do it when you ask. | Free; automatic analysis is billed |
remove_watchlist_channel · delete_watchlist | Stop monitoring a channel, or delete a watchlist and the videos it recorded. | Free |
account | Plan and remaining Cloud/AI balances. | Free |
analyze_video | Frame-level analysis — slides, charts, demos, on-screen text. estimateOnly quotes the price first. | 1 Cloud Request, then AI Units per minute: Standard analysis 4, Deep 30, Quick 2.8 |
list_analyses | Your AI Watch analyses, newest first, so one can be reopened. | Free |
get_analysis | Read a finished analysis; waitSeconds waits for it. | Free |
ask_video | Ask against a finished analysis; citations verified or dropped. | 6 AI Units per question |
Re-reading a video the account already has is free — another time range, another page, or a search after a fetch — and every metered result reports what it charged.
Prefer search_transcript when you have a question. get_transcript costs the same but returns everything, and a long transcript crowds out the rest of your conversation.
If the tools do not appear
- Nothing changed after editing. The app was not fully quit, or the file is not the one being read — check the path for your OS above.
- All servers vanished. Invalid JSON: a duplicate
mcpServerskey, a trailing comma, or a smart quote pasted from a web page. - 403 on every call. Unverified email.
- Windows path escaping. Backslashes inside JSON strings need doubling.