# Every word. Every speaker.

Source: https://corespeed.io/integrations/transcription

Transcription lets Claude Code, Codex, Cursor or any MCP agent turn speech in an audio or video file into text through CoreSpeed, with no account or API key. The transcript marks each speaker and sounds such as laughter, and can add word timings. One call takes a recording of up to 60 minutes.

## Set up an agent

Give your agent this line: `set up https://corespeed.io/SKILL.md (read it raw, e.g. with curl)`

It fetches the skill at https://corespeed.io/SKILL.md, adds the CoreSpeed MCP server, signs you in (or up), and keeps the skill for next time. Nothing to connect: Transcription is built in. Each call is priced in credits, from your one balance. Create an account: https://app.corespeed.io/sign-up

## Send the recording. Get the transcript

Your agent sends the whole recording in one call, so the speakers stay consistent. It then works from the text and saves the transcript where you ask.

You ask your agent: "Transcribe calls/customer-call.m4a from our Drive with speakers and timestamps. List each question the customer asked, and save the transcript next to the call."

It calls CoreSpeed:

- `media__transcribe`: ok, 48200 ms
- `drive__put`: ok, 230 ms

It reads from: Transcription.

### Call transcript: Six questions, in the customer's words.

Speaker 1 is the customer. Each question has the minute it was asked.

**Questions**

- 04:12 Can we cap spend per agent?
- 11:48 Where do the files go?

**Saved**

- The full transcript, next to the call in your Drive.

_Illustrative example · Fictional call and customer_

Transcribe: speech to text: https://corespeed.io/docs/media#transcribe-speech-to-text

## From a recording to the text you need

Requests you can make right after setup. The data is fictional.

### Pull questions from a call

What did the customer ask? · Two speakers

- Recording: A 32-minute customer call
- Found: Six questions, with timestamps
- Speakers kept apart

### Caption a product video

Timed lines for subtitles · Word timings

- Source: A 90-second launch video
- Back: Each word with its start and end
- Ready to cut into captions

Full example: https://corespeed.io/integrations/video-generation (Markdown: https://corespeed.io/integrations/video-generation.md)

### Check a voice-over

Did it say the script? · Speech check

- Script: 40 words
- Heard: The same 40 words
- Generated speech, checked as text

Full example: https://corespeed.io/integrations/fish-audio (Markdown: https://corespeed.io/integrations/fish-audio.md)

## Keep your agent. Nothing to connect

One CoreSpeed account, across compatible agents.

1. **Set up CoreSpeed**: Give your agent the setup line. It adds CoreSpeed and signs you in.
2. **Check the price**: Every tool has a rate in credits. Your agent reads it before each call.
3. **Start with one task**: Ask your agent to transcribe one short recording from your Drive.

Setup guides: Claude Code https://corespeed.io/docs/mcp#claude-code · Codex https://corespeed.io/docs/mcp#codex · Cursor https://corespeed.io/docs/mcp#cursor

## Questions

### Which files can it transcribe?

WAV, MP3, AAC or M4A, FLAC, Ogg, WebM and MOV, up to 60 minutes and 100 MB, from your Drive or a public link.

### Does it tell the speakers apart?

Yes. Each speaker is marked in the text. Send the whole conversation in one call, because labels stay consistent only within one answer.

### Do I need to name the language?

No. The language is detected on its own. You can pass one as a hint.

### Do I need an account or API key for Transcription?

No. Transcription is built into CoreSpeed. Your agent sees its tools after setup, with nothing to connect.

### What does Transcription cost?

Each call has a rate in credits, shown before it runs, from your one balance. A failed call costs nothing. Every Transcription tool and its price: https://corespeed.io/catalog/transcription

### Do I have to switch agents?

No. CoreSpeed adds tools to the agent you already use: Claude Code, Codex, Cursor, or another MCP client. Each one signs in once.

Every tool and price: https://corespeed.io/catalog/transcription

## Related pages

- [Audio generation for AI agents](https://corespeed.io/integrations/audio-generation): Give your agent music, sound effect and text to speech models with no API keys. It reads each rate first, and a failed generation costs nothing.
- [Fish Audio for AI agents](https://corespeed.io/integrations/fish-audio): Fish Audio's voices in your agent, with no Fish Audio account: search the voice library, design a new voice, and speak a script or a whole scene.
- [Media understanding for AI agents](https://corespeed.io/integrations/media-understanding): Let your agent watch a video, look at an image or listen to audio, and answer your question in text. Files from Drive, a link or YouTube.
- [Tools](https://corespeed.io/tools): Models, the live web and nine social platforms, on one balance. No account for any of them.
- [Video generation for AI agents](https://corespeed.io/integrations/video-generation): Give your agent hundreds of text to video and image to video models, with no API keys. It reads the rate first, and long renders finish as jobs.

## Sources

- Media docs: transcribe: https://corespeed.io/docs/media
- Transcription catalog: https://corespeed.io/catalog/transcription
- The media tool now speaks: https://corespeed.io/blog/the-media-tool-now-speaks

Reviewed 2026-11-14.

## Links

- Product overview: https://corespeed.io/ (Markdown: https://corespeed.io/index.md)
- All integrations: https://corespeed.io/integrations (Markdown: https://corespeed.io/integrations.md)
- Documentation: https://corespeed.io/docs
- Pricing: https://corespeed.io/pricing (Markdown: https://corespeed.io/pricing.md)
