New: Import recordings from Google Drive, OneDrive, Zoom, Dropbox & Box. First 3 imports free.

Import a recording
Pepys

Integrations

Connect ChatGPT and Claude to Pepys

Pepys runs an MCP server, so an AI agent can transcribe a file or a pasted link on its own and use the transcript as context.

In short

Pepys runs a hosted MCP (Model Context Protocol) server at https://pepys.co/api/mcp, listed in Anthropic's own Claude Connectors Directory – search "Pepys" inside Claude and add it in one click, no URL to copy. The agent signs in to Pepys once with OAuth – no API key – then transcribes audio, video, and whole podcast feeds and reads the transcript back. You can also run the pepys-mcp server locally with your own API key.

Pepys in Claude's Connectors Directory

Pepys is a listed connector in Claude's own Connectors Directory – it is not just an MCP server you have to find and configure yourself, it is browsable and addable from inside Claude. That listing carries an automated Anthropic review, a fixed connector URL (https://pepys.co/api/mcp), and a published tool list, categorized under Media and entertainment, Design, and Productivity.

To add it: in Claude, open Settings → Connectors → Browse connectors, search "Pepys," and add it. There is no URL to copy and no manual configuration – sign-in is OAuth against your own Pepys account, and usage bills to that account's credits, the same as connecting any other way.

This route also covers Claude in Chrome, which shares Claude's connector list with no extra setup.

Two other ways to connect

Hosted connector – no API key. Point a remote MCP connector at https://pepys.co/api/mcp directly instead of going through the directory. The agent signs in to Pepys once through OAuth. This is the route for ChatGPT, which does not have its own connector directory yet: Settings → Connectors → Add custom connector (Claude has the same option, for anyone who prefers a direct URL over browsing).

Local server – your own key. Run the published pepys-mcp server with a Pepys API key. This suits local dev agents that already have a terminal to run it from.

Local MCP server config
{
  "mcpServers": {
    "pepys": {
      "command": "npx",
      "args": ["-y", "pepys-mcp"],
      "env": { "PEPYS_API_KEY": "pk_live_your_key" }
    }
  }
}

What the agent can do

Both connection methods expose the same tools: transcribe (a file, direct link, or pasted YouTube/podcast URL), upload_file (hand the agent a local file via a presigned upload), get_transcription (read back the transcript, segments, summary, and speaker labels), list_transcriptions, list_podcast_episodes, and export_transcript (correctly-timed SRT, VTT, or plain text).

There are also transcribe_podcast_feed (fan a whole RSS or Apple Podcasts feed out into one transcript per episode), search_transcript (find the moment a phrase was said), ask_transcription (answer a question with citation-backed quotes only), and get_credit_balance.

The same engine behind pepys.co handles the work: automatic language detection across 99+ languages, translation with timestamps preserved, and files from short clips to hours-long recordings.

Example prompts

Once connected, these run directly in the agent's chat.

Try these in the agent
Transcribe https://youtube.com/watch?v=... with speaker labels and give me a summary.

Transcribe the latest 3 episodes of this podcast feed and tell me how
much of my Pepys balance that used: https://feeds.example.com/show.rss

In my last transcription, find every place the speakers mention "pricing"
and quote the exact timestamps.

Export my most recent transcript as an SRT file.

What it costs

The 60 free minutes granted when you create an account apply through the MCP too. After that it is pay-once and usage-based – 1 credit = 1 minute, credits never expire, no subscription. See pricing for the current rates.

Diarization (speaker labels), whole-feed batch transcription, and word-level export unlock with any one-time purchase, the same as everywhere else in Pepys.

Also published in

Beyond Claude's own directory, Pepys is published in the other places agents and MCP tooling browse: Smithery, mcp.so, and the MCP Registry.

The full REST API behind both connection methods is documented from authentication onwards, including a dedicated MCP article, and the MCP server page is the short overview.

Connect ChatGPT and Claude to Pepys – questions, answered

Is Pepys listed in Claude's Connectors Directory?

Yes. Search "Pepys" in Claude's Settings → Connectors → Browse connectors and add it directly – no URL to copy. It is a Community-tier listing with an automated Anthropic review, at https://claude.ai/directory/pepys-co.

Does Pepys have an MCP server?

Yes. Pepys runs a hosted MCP server at https://pepys.co/api/mcp that any MCP client can connect to, and there is a published pepys-mcp server you can run locally with your own API key.

How do I connect Claude or ChatGPT to Pepys?

In Claude, the fastest way is Settings → Connectors → Browse connectors, then search "Pepys." Otherwise, add a custom connector pointing at https://pepys.co/api/mcp – this is the only route for ChatGPT today. Claude in Chrome uses Claude's own connector list. You sign in to Pepys once with OAuth.

Do I need an API key for the MCP server?

Not for the hosted connector – it uses OAuth sign-in. The local pepys-mcp server uses a bearer API key from Settings → API keys instead.

What can an AI agent do with the Pepys MCP?

Transcribe a file or a pasted link, transcribe a whole podcast feed, get speaker diarization, export correctly-timed SRT or VTT, search across a transcript, and check the credit balance without leaving the agent.

What does the MCP cost?

The 60 free minutes from signup apply through the MCP too. After that it is pay-once and usage-based: 1 credit = 1 minute and credits never expire. Diarization, whole-feed batch, and word-level export unlock with any one-time purchase.

Related

Try it on your own audio

Create a free account and transcribe a recording – 60 minutes are included, no card required. Still stuck? Email contact@pepys.co or see the support page.