# TranscriptLab MCP server (transcriptlab.dev)

You are helping a user connect this MCP server to their AI client. If you are a coding agent that can change your own configuration (Claude Code, Cursor, Codex and similar), add the server yourself using the matching section below, then tell the user if a restart or reload is needed. The first tool call opens a TranscriptLab consent page where the user signs in or pastes an API key; you cannot complete that step for them. If the user has not given you an API key, ask for it. Never invent one.

## Server

- URL: https://api.transcriptlab.dev/mcp
- Transport: streamable HTTP, remote. No local install.
- Authentication: OAuth 2.0 with dynamic client registration (RFC 7591). Connecting and listing tools needs no sign-in. The first tool call opens the TranscriptLab consent page: the user signs in with their TranscriptLab account or pastes a TranscriptLab API key.
- Access tokens are scoped transcriptlab:read and transcriptlab:write, last one hour, and refresh automatically.
- A connection made with an API key can use every tool except the session-only ones: transcriptlab_export_link (download links) and large-file uploads.
- Accounts and keys: https://transcriptlab.dev (dashboard). Support: hello@transcriptlab.dev.

## Tools

- transcriptlab_get_transcript: Get the transcript of a YouTube video from its captions, optionally in another language. Cost: First delivery of a variant can use a credit.
- transcriptlab_retrieve_transcript: Read a YouTube transcript that is already published for your account, without starting a new fetch. Cost: Free for variants you already have.
- transcriptlab_discover_captions: List the caption languages and track types (manual or generated) a video offers. Cost: Free, subject to a monthly limit.
- transcriptlab_get_job_status: Check whether a transcript job has finished. Cost: Free.
- transcriptlab_search_transcript: Search, quote or page through a transcript you already have (keyword, regex or sequential chunks). Cost: Free.
- transcriptlab_list_transcripts: List the transcripts in your library, newest first. Cost: Free.
- transcriptlab_transcribe_file: Transcribe an audio or video file attached to the conversation with speech-to-text. Cost: Credits per audio minute.
- transcriptlab_transcribe_media_url: Transcribe a public https link to an audio or video file. Cost: Credits per audio minute.
- transcriptlab_get_speech_transcript: Read the speech-to-text result for a file or media link you submitted. Cost: Free.
- transcriptlab_export_link: Create a 10-minute download link (TXT, SRT, VTT, Markdown or JSON) for a transcript you have. Cost: Free; needs a signed-in connection.
- transcriptlab_get_usage: Show your plan, remaining credits and usage this period. Cost: Free.
- transcriptlab_open: Open the interactive TranscriptLab view: paste a link, upload a file, browse your library. Cost: Free.
- transcriptlab_open_audio: Entry point a host uses when you open an audio or video file with the connector. You do not call it yourself. Cost: Credits per audio minute.

Key parameters (full schemas come from tools/list):

- transcriptlab_get_transcript: url (required, YouTube URL or 11-character video ID), lang, translate, type (any|manual|generated), format (json|text|srt|vtt), metadata, also (extra formats in a JSON response), cleanup. Returns the transcript, or a job_id when the fetch outlives the synchronous window.
- transcriptlab_retrieve_transcript: url (required), plus the same variant selectors. Reads an already delivered variant; free for variants the account already has.
- transcriptlab_discover_captions: url (required). Lists caption languages and track types.
- transcriptlab_search_transcript: url and mode (required; keyword|regex|chunk). terms in keyword mode, pattern and flags in regex mode, target_tokens/max_tokens in chunk mode. Only works on a transcript already delivered to the account.
- transcriptlab_get_job_status: job_id (required). Poll after get_transcript returns a job; when done, read with retrieve_transcript.
- transcriptlab_transcribe_file: file (required, the attached media file), diarize, tier (normal|premium|auto). Returns upload_id; poll get_speech_transcript. Needs a signed-in connection.
- transcriptlab_transcribe_media_url: url (required, public https link to an audio or video file, not a web page), diarize, tier. Returns media_id; poll get_speech_transcript.
- transcriptlab_get_speech_transcript: exactly one of upload_id or media_id, format. While pending, wait a few seconds between tries and stop after about ten.
- transcriptlab_export_link: url or upload_id, format (text|srt|vtt|md|json), lang. Returns a 10-minute download link. Needs a signed-in connection.
- transcriptlab_get_usage: no parameters. Plan, remaining credits, current-period usage.
- transcriptlab_list_transcripts: search, limit, cursor. Library listing, newest first.
- transcriptlab_open: tab (link|upload|library). Opens the interactive view.

Host-facing tools the model should not call: transcriptlab_mentions, transcriptlab_open_audio, transcriptlab_begin_audio_upload, transcriptlab_start_audio_transcription, transcriptlab_read_audio_transcript, transcriptlab_submit_batch, transcriptlab_list_job_items.

A YouTube video without captions is not transcribed automatically: the tool says so, and speech-to-text is an explicit opt-in with an attached file or a direct media link.

## Claude (web and desktop)

Open Settings, then Connectors, choose Add custom connector, and paste the URL. The first tool call opens the TranscriptLab consent page.

    https://api.transcriptlab.dev/mcp

Agent instructions: The user adds this in the Claude app; you cannot do it for them. Walk them through it: Settings, then Connectors, then Add custom connector. Name it TranscriptLab and set the MCP server URL to https://api.transcriptlab.dev/mcp. The first tool call opens a TranscriptLab consent page where they sign in or paste an API key.
## Claude Code

Run the command in a terminal. The first tool call opens the consent page in a browser.

    claude mcp add --transport http transcriptlab https://api.transcriptlab.dev/mcp

Agent instructions: Add the server yourself by running: claude mcp add --transport http transcriptlab https://api.transcriptlab.dev/mcp (add --scope user to make it available in every project). Then tell the user the first tool call opens a browser consent page where they sign in to TranscriptLab or paste an API key. You cannot complete that step for them.
## Cursor

Add to ~/.cursor/mcp.json, or .cursor/mcp.json inside one project.

    {
      "mcpServers": {
        "transcriptlab": {
          "url": "https://api.transcriptlab.dev/mcp"
        }
      }
    }

Agent instructions: Add this to ~/.cursor/mcp.json, or to .cursor/mcp.json in one project, keeping any servers already listed: {"mcpServers":{"transcriptlab":{"url":"https://api.transcriptlab.dev/mcp"}}}. Tell the user to open Cursor MCP settings and connect transcriptlab if no browser opens by itself; they sign in on the TranscriptLab consent page.
## Codex

Add to ~/.codex/config.toml.

    [mcp_servers.transcriptlab]
    url = "https://api.transcriptlab.dev/mcp"

Agent instructions: Add this to ~/.codex/config.toml, keeping anything already in the file: [mcp_servers.transcriptlab] with url = "https://api.transcriptlab.dev/mcp". Tell the user the first tool call opens a browser consent page where they sign in to TranscriptLab or paste an API key.
## ChatGPT

Open Plugins, choose Add, then Add custom MCP server. Paste the URL and create the plugin. The first tool call opens the consent page.

    https://api.transcriptlab.dev/mcp

Agent instructions: The user adds this in ChatGPT; you cannot do it for them. Walk them through it: Plugins in the sidebar, then Add, then Add custom MCP server. Name it TranscriptLab, set the server URL to https://api.transcriptlab.dev/mcp, tick the confirmation box and create the plugin. The first tool call opens the TranscriptLab consent page.
## Any other MCP client

Point the client at the URL as a remote streamable HTTP server. No client ID is needed. If the client cannot run OAuth, paste a TranscriptLab API key on the consent page; API-key connections cannot use download links or large-file uploads.

    https://api.transcriptlab.dev/mcp

Agent instructions: Configure the client with the remote streamable HTTP server URL https://api.transcriptlab.dev/mcp. It discovers the OAuth endpoints and registers itself (RFC 7591), so no client ID is needed. If the client cannot run an OAuth flow, ask the user for a TranscriptLab API key and have them paste it on the consent page. Never invent a key. API-key connections cannot use the session-only tools (download links, large-file uploads).

## Credits and limits

- A YouTube transcript uses 1 credit on first delivery of a variant; later deliveries of that variant are free.
- Searching, quoting, listing and exporting a transcript you already have costs no credits.
- Speech-to-text: 1 credit per audio minute on the standard model; the premium model (speaker labels) uses 10.
- Caption discovery is free but limited each month by plan.
- Free account: 10 one-time credits. Starter: $3.99/month for 1,000 credits. One-time pack: $4.99 for 1,000 credits.
- Files attached in chat can be up to 300 MB; larger files up to 2 GB go through the upload view. Media links must be public https URLs.
- When credits run out or a limit is hit, the tool says so and stops. It does not retry.

## Check that it works

Call transcriptlab_discover_captions with url set to dQw4w9WgXcQ. It is free and should return the caption languages the video offers. If the result asks the user to connect their account, pass that message on as it is; do not retry in a loop.

Calling the HTTP API from code instead of MCP: https://transcriptlab.dev/api.md
Full reference: https://transcriptlab.dev/docs/mcp
