Audivo

What Audivo MCP server does

Audivo MCP server gives an MCP client access to podcast discovery, transcription, and transcript-management workflows through the Audivo API. It can search shows by name, host, or topic; display category charts; list a show's newest episodes; and retrieve transcripts in pages of about 40,000 characters. The service is suited to agents that need to inspect podcast episodes without loading an entire transcript into one response.

The primary transcribe tool accepts an Apple Podcasts link, a feed URL and GUID, an episode identifier returned by list_episodes, or an existing upload identifier. The local server can also process an absolute local file path or a YouTube link. Previously transcribed episodes return immediately; new requests create a job and may return a job ID if processing continues after the initial wait.

How it works

Audivo MCP server is available in two forms. The hosted service uses Streamable HTTP at https://api.audivo.dev/mcp and supports OAuth or a bearer key. The local package runs through stdio with npx -y @audivo/mcp@latest and reads its API key from the client environment. The hosted mode does not provide YouTube processing; local mode downloads YouTube audio on the machine with yt-dlp.

A single transcription waits briefly for completion. If it is still running, read_transcript can wait for the remaining work and return the result page by page. Each page identifies how to request the next page. Repeating the same transcription request is idempotent and does not create a second charge. The max_credits option can reject a request before spending credits when its estimated cost is too high.

For multiple episodes, use quote to estimate the selection, then confirm to create a job group. group_status, list_groups, and cancel_group provide lifecycle controls. Credits are held up to the quote ceiling and settle against measured audio duration.

Setup and configuration

Create an API key in the Audivo dashboard. For local use, provide it as AUDIVO_API_KEY when configuring the MCP client. Optional settings include AUDIVO_API_BASE_URL for a public HTTPS API origin, AUDIVO_YTDLP_PATH to select a specific yt-dlp binary, and AUDIVO_CACHE_DIR for the managed download cache.

The local server can be added to Claude Code, Codex, Cursor, Claude Desktop, Windsurf, VS Code, and other clients that launch stdio processes. Keep configuration containing a live key out of source control and shared messages. The hosted endpoint is appropriate for clients that accept an MCP URL.

Tools and capabilities

Audivo MCP server exposes tools for:

• Searching shows, viewing category charts, and listing episodes.
• Transcribing podcast episodes and reading completed transcripts by page.
• Listing recent transcripts and transcript titles.
• Estimating and confirming multi-episode work.
• Checking, listing, and cancelling job groups.
• Uploading local audio and finding YouTube episodes in local mode.

Clients that support MCP Apps can display an embedded transcript reader and transcript library. The reader includes timestamps, transcript search, episode details, and on-demand pages. Clients without MCP Apps receive the regular tool results instead.

Limitations and notes

Transcription consumes credits, at about one credit per audio minute, while discovery and reading tools do not. The confirm operation can spend credits up to the quoted ceiling. Hosted transcription waits up to its gateway limit, while the local server has a longer default wait and can report progress to clients that request it.

YouTube search and local file upload are local-server features. The hosted endpoint does not download YouTube media. MCP Apps support also depends on the client; clients without that extension ignore the embedded interface and use text results.