You're already pasting transcripts into Claude to summarise an interview, pull quotes from a YouTube video, or turn a podcast into notes. The trip out to a transcription tool and back is a small annoyance that adds up over a week.
Here's how to skip the round trip. Once VTS is connected to Claude AI, you hand Claude a video link inside the chat and the transcript comes back in the same conversation, ready for follow-up questions, summaries, or rewrites. The setup is the same three steps whether you live in Claude.ai (the browser) or Claude Desktop — both use the custom-connector flow with OAuth, so there's no API key to copy around.
- Three steps, same flow on Claude.ai web and Claude Desktop — open Settings → Connectors, paste the VTS URL, approve the OAuth sign-in. Takes about a minute.
- No API keys, no config files, no Python. OAuth handles authentication; tokens refresh silently.
- After setup, say something like "transcribe https://youtube.com/…" in any chat and Claude calls VTS automatically.
- Free tier — you get 300 minutes of transcription per month without paying. Past that, top up your wallet (6¢/min PAYG) or subscribe to Pro for 25 hours/month included.
- Works with songs and noisy audio too — the auto-escalation ladder tries the heavier engines automatically when the first pass comes back rough. See /sla for the accuracy bands.
What you can do with VTS inside Claude
Once the connector is wired in, eight tools become available to Claude — none of them require you to leave the chat or remember a syntax:
- Transcribe a YouTube link — "What does this video say? https://youtube.com/watch?v=…" — Claude pastes the transcript and starts answering questions about it.
- Transcribe a Facebook video or reel — same flow, same URL pattern.
- Pick a format — Claude asks "plain text, timestamps, or SRT subtitles?" and fetches the chosen format.
- Search your past transcripts — "find the transcript where the founder mentioned pricing" — VTS returns matches from your saved transcript library.
- Get speaker labels — opt in to diarisation when there's more than one voice.
The Apps SDK widgets render the result as a clickable transcript card with the video thumbnail, format picker, and download links — same UI on mobile and web. Nothing to install for that; it shows up automatically once you've added the connector.
Before you start
Short list — both paths share the same prereqs because they use the same connector flow.
You'll need a VTS account. If you don't have one yet, paste any short YouTube link at vts.askgiya.com — it'll prompt you to sign in via magic link the first time you need paid features (or 300 free minutes/month, whichever comes first).
- A Claude.ai account (free or paid both work).
- For the web path: any modern browser.
- For the Desktop path: Claude Desktop installed (macOS or Windows).
That's it. No API key, no Python, no config files.
Option 1: Add VTS to Claude.ai (browser)
About 60 seconds end to end. Claude handles authentication for you via OAuth — no API key to copy around.
Click your avatar → Settings → Connectors in the sidebar. Expected result: you see a list of connectors (Google Drive, Gmail, etc.) and an Add custom connector button at the bottom.
Click Add custom connector, enter Server URL https://vts.askgiya.com/mcp and Name VTS Transcription, then click Add. Expected result: a new connector card appears in your list with a Sign in link.
Click Sign in — a popup opens pointing at https://vts.askgiya.com/oauth/authorize?…. Sign in with your VTS magic-link email if needed, click Approve, and the popup closes. Open any chat and paste: Transcribe this and summarise the three main points: https://www.youtube.com/watch?v=…. Expected result: Claude shows "Using VTS Transcription…" briefly, returns a transcript card with the video thumbnail and a Plain/Timestamps/SRT picker, then continues with the summary.
Common failure → fix:
- "Sign in to enable" never goes away. The OAuth popup might have been blocked. Open Claude's site permissions and allow popups from claude.ai, then click the connector again.
- Tools don't show up in the picker. Refresh the chat or start a new one — connectors load when a chat session starts, not retroactively.
- Sign-in popup every hour. Shouldn't happen — VTS issues refresh tokens so the connector stays signed in for 60 days. If you're seeing it, your client predates refresh-token support; restart it or update to the latest Claude/ChatGPT build.
Option 2: Add VTS to Claude Desktop
Pick this path if you prefer the native app over the browser. The flow is the same three steps as the web path — same connector URL, same OAuth, same approval screen.
Click your name or the gear icon in the sidebar → Settings → Connectors. Expected result: a list of available connectors with an Add custom connector button at the bottom.
Click Add custom connector, then enter Server URL https://vts.askgiya.com/mcp and Name VTS Transcription. Click Add. Expected result: the connector card appears in your list with a Sign in link.
Click Sign in — Claude Desktop opens the VTS authorization page. Sign in with your VTS magic-link email if you're not already in, click Approve, and the popup closes. Start a chat: Transcribe this and pull the three best quotes: https://www.youtube.com/watch?v=…. Expected result: Claude shows "Using VTS Transcription…", returns the transcript card with thumbnail + format picker, then answers your question.
Common failure → fix:
- "Add custom connector" button isn't there. Custom connectors require a recent Claude Desktop build — open About and make sure you're on the current release, then restart.
- Sign-in popup closes immediately. Pop-ups blocked, or the system browser intercepted the redirect. Try the connector again — Claude Desktop falls back to its in-app browser on the second attempt.
- Sign-in popup every hour. Shouldn't happen — VTS issues refresh tokens so the connector stays signed in for 60 days. Access tokens rotate hourly behind the scenes; you only see a popup once, after consent.
What to try once it's connected
Real-world prompts that work well in both paths:
- "Pull the five best quotes from this interview, with timestamps." — Claude calls
transcribe_url, thenget_formatfor the timestamped version, then formats the quotes inline. - "This podcast just dropped, write me a LinkedIn post about its main argument: [url]" — Transcript → analysis → post draft, all in one turn.
- "I have a Spanish-language Facebook reel about election fraud at [url]. Transcribe it and translate the key claims to English." — Whisper detects Spanish automatically; Claude does the translation.
- "Find the interview transcript in my library where Lisa talked about the Tuesday meeting." — Claude calls
search(query="Lisa Tuesday meeting"), thenfetch(id=…)to pull the full text.
The pattern is the same across all of them: paste the video URL in your prompt, ask the question you actually want answered, and Claude figures out which VTS tools to call. You don't have to name the tools.
Common setup issues (and fixes)
A handful of issues come up often enough to call out:
OAuth pop-up shows "404 Not Found". This means your Claude.ai connector is pointing at the wrong path. Double-check the server URL is exactly https://vts.askgiya.com/mcp — no trailing slash, no /api prefix. Some users paste the homepage by mistake.
Free tier ran out mid-transcription. Anonymous and free-tier accounts get 300 minutes per month with a 60-minute daily cap and a 20-minute per-file cap. When you hit any of them, VTS returns a structured error and Claude relays it as "you've used your free minutes for the month, top up your wallet to continue". Just go to vts.askgiya.com/billing and add credit — your next prompt in the same chat picks up where it left off.
Long files take a few minutes. Audio over 30 minutes can take 1-3 minutes to transcribe even on the medium model. Claude shows "Transcribing…" while it waits — don't refresh, you'll lose the chat context. For files over an hour, consider splitting them at natural breaks for higher per-segment accuracy.
Speakers aren't labelled. Diarisation is opt-in — say "transcribe with speaker labels" in your prompt and Claude passes diarize=True to the tool. Note the per-file cap on diarised audio is shorter (20 minutes on CPU), so very long meetings won't get labels.
The transcript looks wrong on a song. Songs are hard — see /sla for the expected accuracy bands by audio type. After the first pass, Claude can re-run with denoise via "re-transcribe this with noise reduction" — VTS auto-escalates through a 3-tier ladder (small → medium with denoise → OpenAI large) until accuracy clears the floor for the audio type.
FAQ
Does it cost anything to add the connector?
No. Adding the connector is free. You pay the same way you'd pay using VTS directly — 300 minutes free per month, then 6¢/min from your wallet, or 25 hours/month included on the Pro plan. Claude doesn't add a markup on top.
Is my data sent to Claude or Anthropic?
The transcript Claude receives flows through Claude's normal chat history — so anything you do with it (summarise, quote, paste into a document) is subject to Anthropic's data handling. The audio file itself is never sent to Anthropic; VTS extracts the audio, transcribes it on our infrastructure, and only returns the text + a download link.
Can I use the same connector for ChatGPT?
Yes. The same MCP URL (https://vts.askgiya.com/mcp) works as a custom connector in ChatGPT (web and desktop), and ChatGPT's Deep Research mode picks up VTS as a searchable source via the search and fetch tools. The OAuth flow is identical.
What happens to my transcripts after the chat ends?
Saved transcripts live in your VTS account at vts.askgiya.com/dashboard — they don't disappear when the Claude chat does. You can search them later from any tool, or download them as .txt / .srt files.
Can Claude transcribe live audio (microphone input)?
Not yet. Both paths work with pre-recorded video URLs (YouTube, Facebook, and a few others). Real-time transcription is a different shape of API; if it'd be useful for your workflow, let us know.
Paste any public link or upload a file and get a clean transcript in minutes. First 3 clips every month are on us — no card required.
Once you've got VTS wired into Claude, the natural next move is to pair it with one of the longer-form workflows — transcribing a research interview for a paper, pulling show notes from a podcast for content repurposing, or comparing the free transcription tools to know when paid is worth it. Claude turns each of those into a one-prompt task instead of a five-tab juggle.
Sources
- Model Context Protocol — official spec — the open standard VTS implements on the server side.
- Claude Desktop download — official download page.
- Anthropic — custom connectors documentation — Claude's own walkthrough of the connectors UI and OAuth flow.
- VTS accuracy expectations — published bands by audio type, plus the auto-escalation ladder VTS runs to honour them.
