Scriptivox

    MCP server

    Scriptivox runs an official Model Context Protocol server, so Claude, ChatGPT, Cursor, Codex and any other MCP client can transcribe audio and video, and work with your transcripts, as native tool calls.

    There is one address for every app:

    https://platform.scriptivox.com/mcp

    Add it, sign in once in the browser, and choose what the app uses. Nothing to install.


    Connect an app

    Claude (web, desktop and mobile): Scriptivox is in Claude's connector directory, verified by Anthropic. Open Scriptivox in Claude's connector directory, check that it says Anthropic verified and Made by Scriptivox, click Add to Claude and sign in. Or open Customize → Connectors, choose + Add → Add custom connector, name it Scriptivox, paste https://platform.scriptivox.com/mcp and sign in.

    ChatGPT: on the web, open Plugins, choose + → Add custom MCP server, name it Scriptivox, paste https://platform.scriptivox.com/mcp, choose OAuth, then Create as a plugin and sign in.

    Claude Code

    bash
    claude mcp add --transport http scriptivox https://platform.scriptivox.com/mcp

    Then run /mcp inside Claude Code and choose Authenticate.

    Codex

    bash
    codex mcp add scriptivox --url https://platform.scriptivox.com/mcp
    codex mcp login scriptivox

    One-click: Add to Cursor · Add to VS Code

    Any other client. Add it as a remote server:

    json
    {
    "mcpServers": {
    "scriptivox": {
    "url": "https://platform.scriptivox.com/mcp"
    }
    }
    }

    A client that can only run local (stdio) servers can bridge to it with npx mcp-remote https://platform.scriptivox.com/mcp.

    What you choose when you sign in

    The sign-in page asks two things. You can change both later on Connected apps.

    1. What the app uses

    My Scriptivox planAPI credits
    BillingYour plan's minutes and limits, exactly like the websiteYour prepaid API balance, $0.20 per hour of audio
    ForMost people: ChatGPT, Claude, everyday useDevelopers and scripts
    AI-app allowanceOn every paid plan, AI apps can start 100 transcriptions a day and run 10 at once (the website itself has no daily limit). For more, switch the connection to API creditsThe API's own limits and your balance
    BuyingCannot buy anything or create API keys; at a limit it points you to pricingCan ask to buy a plan, add credit or manage API keys, each only after you approve it
    FilesUpload box in the chat (up to 5 GB), ChatGPT attachments, links, upload links for apps with a terminalUpload box in the chat (up to 5 GB), links, upload links for apps with a terminal
    LinksGoogle Drive, Dropbox, OneDrive, SharePoint and Teams recordings (shared with "Anyone with the link"), TikTok, Instagram, Facebook, X, Snapchat, and direct links to an audio or video file on any public website. Not YouTube: send the video as a file insteadGoogle Drive, Dropbox and OneDrive sharing links, public TikTok, Instagram, Facebook, X and Snapchat posts, and direct links to an audio or video file

    The plan is preselected. Someone with API credits and no paid plan gets API credits preselected.

    2. What it may do

    • Everything (default): transcribe, read, organise and edit, send meeting bots, email summaries and run automations. The app must ask you in the chat before anything that reaches other people.
    • Transcribe and organise: everything except meeting bots, emails, automations and purchases.
    • Read only: find, read and export your transcripts. Nothing is transcribed, changed or sent.

    What it can do in your library

    • Delete: a deleted transcript goes to Trash on the website, as when you delete it there; only you can restore it, on the website, within 30 days. The app cannot see, restore or empty Trash. Deleting a folder sends the transcripts in it to Trash too. A transcription that is still running cannot be deleted.
    • Organise: create, rename and delete folders and workspaces; move and tag transcripts.
    • Export: PDF, Word, TXT, CSV, SRT or VTT, several at once as one ZIP, with the website's export options (words, characters and seconds per caption, sentence breaks, characters to remove, timestamps, speakers). A text format comes back in the chat; files come as a download card.
    • Calendar: if you connected a calendar on the website, the app lists your upcoming meetings and turns the recording bot on or off for each, or for all. Connecting or disconnecting a calendar is only done on the website.
    • Asks you first: the app asks in the chat before deleting a folder, a workspace or several transcripts at once, switching the bot for several meetings or "record all meetings", sending a meeting bot, emailing a summary or running an automation.
    • Website only: Trash, creating or editing automations, deleting a tag from your account, connecting a calendar, AI chat and the Summary tab.
    • In ChatGPT: the same tools, except automations and anything about plans or prices (ChatGPT's rules for apps).

    If you are an AI assistant asked to install Scriptivox

    1. Add the server https://platform.scriptivox.com/mcp with the command for your client above.
    2. Ask the person to finish signing in (in Claude Code: /mcp → Authenticate). A browser opens; they choose their plan or API credits and approve. Do not ask them for an API key or a password.
    3. Check it worked: call get_account, then tell the person which account you are connected to and what this connection uses.
    4. If a call fails, look the message up in Troubleshooting and tell the person what to do. Do not retry the same failing call in a loop.

    Files on the person's computer

    The server cannot read anyone's disk. There are three ways in, best first:

    • In a chat app: call upload_file and the person drops the file in the upload box (both kinds of connection; on API credits each file is billed to the API balance). In ChatGPT a file attached to the message is picked up by transcribe_file. Claude cannot hand a chat attachment to a connector, so in Claude chat the app calls upload_file with the attached file's name, and the box asks the person to pick that same file. In Claude the box also shows how many transcriptions are left today on a plan with a daily limit, and with none left it says when the count starts again instead of showing a file picker.
    • An app with a terminal (Claude Code, Codex, Cursor): call upload_link with the file's name. It returns a one-time upload address and the exact curl command to run. Once the upload finishes, the app calls start_transcription (like pressing Transcribe on the website); nothing is transcribed or counted before that, so an upload that never arrives leaves nothing behind.
    • A link: transcribe_url with a share link.

    Your controls

    On Connected apps. Only you, signed in on the website, can change these; a connected app cannot, even though it acts for your account.

    • What each app uses and may do. Change either for any connection.
    • Activity. Each connection has Show activity: every tool it called, when, and whether it worked. Never the content. Kept for 90 days.
    • Let AI apps use my account. Switch it off to block every connected app at once. Nothing is disconnected; switch it back on and they carry on.
    • Ask me before an app spends money or changes API keys. On by default. The app gets an APPROVAL NEEDED link to platform.scriptivox.com/approvals/<id> and nothing happens until you approve. An approval works once, only for exactly what was asked, and expires after 24 hours.
    • Ask me before an app sends a meeting bot or runs an automation. Off by default; works the same way.
    • Emailing a meeting summary to someone new always waits for your approval, whatever the switches say.
    • Limits. 120 requests a minute per connection, 300 per person (429 with Retry-After past that); at most 5 meeting bots an hour; at most 10 connected apps. On a plan connection, AI apps can start 100 transcriptions a day (reset at 00:00 UTC) and run 10 at once on every paid plan; past that the app says so and suggests API credits.
    • Expiry. A connection unused for 90 days ends, and every connection ends a year after you approved it. Sign in again to reconnect.

    Every sign-in is its own connection: Claude Code on two computers shows up twice. The sign-in page suggests a name such as "Claude Code on Linux"; each connection also has a short id like mcp_3fa9c1d2 and shows the app's software and where it signed in from.

    Troubleshooting

    What you seeWhat it meansWhat to do
    The client says the server needs authenticationYou have not signed in yet, or the sign-in endedIn Claude Code run /mcp → Authenticate; in Codex codex mcp login scriptivox; in Claude or ChatGPT reconnect the connector
    401 invalid_token with "revoked", "expired" or "more than a year old"It was disconnected, your password changed, it went unused for 90 days, or it is a year oldSign in again; that makes a new connection
    "…is not available on a connection that uses the Scriptivox plan"That tool needs API creditsChange the connection to API credits on Connected apps, or reconnect and choose it
    "…is only available on a connection that uses the Scriptivox plan"The upload box and ChatGPT attachments use the planChange the connection to your plan, or use upload_link or a link
    "…was approved as Read only"The access level leaves that tool outChange the access level on Connected apps
    403 with "already has 10 connected apps"The per-account limitDisconnect one, then sign in again
    403 with "AI app access is turned off"The owner switched AI apps offSwitch Let AI apps use my account back on
    429 with "Too many requests"Over the rate limitWait for the Retry-After seconds
    APPROVAL NEEDEDAn action waits for the account ownerOpen the link, approve, then call again with the same arguments plus approval_id
    ZERO_BALANCE or INSUFFICIENT_BALANCEAn API-credits connection has no balanceAdd credit, or switch the connection to your plan
    The sign-in page shows a red warningThe app's return address is not Scriptivox, not this computer and not a known app, or its name claims to be ScriptivoxCancel unless you started this sign-in yourself just now
    Anything else, or an error you can't explainNot covered aboveEmail support@scriptivox.com with the connection id, the job id if there is one, and the time. It is resolved within 24 hours, our founders' personal guarantee

    API keys

    Scripts and clients that cannot sign in can send a key instead. A key always uses API credits, and covers transcription and exports; the account tools (library, meetings, automations, buying) need a signed-in app.

    json
    {
    "mcpServers": {
    "scriptivox": {
    "url": "https://platform.scriptivox.com/mcp",
    "headers": {
    "Authorization": "Bearer sk_live_YOUR_KEY"
    }
    }
    }
    }

    Authorization: Bearer sk_live_…, bare Authorization: sk_live_…, or X-Api-Key: sk_live_… all work. Get one at /keys; see Authentication. Calls made with a key show in that key's activity on /keys.

    Tools

    72 tools. A plan connection sees 58 of them, an API-credits connection or key 70; the access level chosen at sign-in narrows that further ("lowest level" below). Generated from the server's own tool list.

    ToolPlanAPI creditsLowest levelWhat it does
    get_supported_languages✓✓Read onlyList all languages supported by Scriptivox for audio/video transcription.
    get_pricing✓✓Read onlyGet Scriptivox pricing: the web plans (subscriptions and lifetime deals) with their prices and what each includes, read live from the plan list behind the pricing page, and the pay-as-you-go API rate.
    get_product_info✓✓Read onlyGet information about Scriptivox capabilities: transcription, audio tools, video tools, subtitle tools, meeting bot, or API.
    get_api_docs✓Read onlyGet Scriptivox API documentation.
    check_balance✓Read onlyCheck your Scriptivox API credit balance, available hours, and pricing.
    transcribe_url✓✓Transcribe and organiseTranscribe audio or video from a public URL using Scriptivox AI.
    transcribe_file✓Transcribe and organiseTranscribe an audio or video file the person attached to their message in this chat.
    transcribe_status✓✓Read onlyCheck the status of a Scriptivox transcription job.
    transcribe_upload✓Transcribe and organiseStart transcribing a file you already sent with upload_link: pass its upload_id.
    upload_link✓✓Transcribe and organiseSend a file from the computer you are running on to Scriptivox, when you can run terminal commands (Claude Code, Codex, Cursor): returns a one-time upload address and the exact curl command to run.
    start_transcription✓Transcribe and organiseStart transcribing a file that is waiting for it (one you sent with upload_link, once its command finished without an error, or one the person uploaded in the upload box, when they ask you to start it): like pressing Transcribe on the Scriptivox website.
    transcribe_cancel✓Transcribe and organise · can overwrite or removeCancel an in-flight Scriptivox transcription and release any reserved balance.
    transcribe_delete✓Transcribe and organise · can overwrite or removeSoft-delete a Scriptivox transcription record.
    list_transcriptions✓Read onlyList recent API transcriptions for the signed-in account, with optional status/date filters and cursor pagination.
    export_transcript✓✓Read onlyExport a completed Scriptivox transcript as SRT subtitles, WebVTT subtitles, or plain text.
    get_account✓✓Read onlyRead the account of the signed-in person: current plan, entitlements and quota.
    get_credits✓✓Read onlyCheck what is left on the signed-in person's Scriptivox plan right now: AI credits (and when they refill), transcriptions left today, transcription minutes left this month, meeting-bot minutes left, and storage used.
    get_email_settings✓✓Read onlyShow which Scriptivox emails the signed-in person gets and which ones they can turn off: activity emails (transcripts ready, daily limit resets, meeting summaries), tips and offers (marketing), and notes from the team, each on or off.
    set_email_settings✓✓Transcribe and organiseTurn the person's Scriptivox email switches off or on, the same switches as Settings on the website and the Unsubscribe link in each email.
    create_api_key✓EverythingMint a new Scriptivox API key (sk_live_...) for the signed-in person.
    revoke_api_key✓Everything · can overwrite or removePermanently revoke an API key belonging to the signed-in person.
    purchase_plan✓EverythingStart checkout for a Scriptivox web subscription and return a Stripe Checkout URL.
    top_up_balance✓EverythingStart checkout to add credit to the prepaid API balance and return a Stripe Checkout URL.
    get_billing_history✓✓Read onlyRead the signed-in person's billing history: plan invoices, add-on charges, lifetime purchases and API deposits, newest first, with links to each Stripe invoice.
    get_billing_portal_url✓EverythingReturn a link to the Stripe billing portal for the signed-in person, where they can change their card, download invoices, or cancel a subscription.
    search_transcripts✓✓Read onlyFind transcripts in the signed-in person's library by FILE NAME, folder, tag, workspace, status, date, source (upload, link, meeting…), language or length, sorted as asked (newest first by default).
    find_in_transcripts✓✓Read onlySearch what was SAID across every transcript in the signed-in person's library: which recordings mention a word or phrase, ranked, each with its matching lines (timestamp, speaker, segment number).
    rename_transcription✓✓Transcribe and organiseRename a transcript in the signed-in person's library (its title, shown everywhere on the website).
    rename_speakers✓✓Transcribe and organiseGive the speakers in a transcript real names ("SPEAKER_00 is Priya"), the same as renaming speakers on the website: every view, export, share link and get_transcript then shows the names.
    edit_transcript✓✓Transcribe and organise · can overwrite or removeCorrect the text of a transcript, the same as editing it on the website: the edited text becomes the transcript everywhere (viewer, exports, share links, chat), and word timings are re-aligned.
    restore_original_transcript✓✓Transcribe and organise · can overwrite or removeThrow away every text edit on a transcript and go back to the original, exactly as it was first transcribed (the website's "Restore original").
    send_transcript_feedback✓✓Read onlySend the person's feedback on a transcript, or a problem report, to the Scriptivox team.
    remove_tags✓✓Transcribe and organiseRemove tags from up to 50 transcriptions (tag_transcriptions adds them).
    move_to_workspace✓✓Transcribe and organiseMove up to 50 transcriptions into another of the person's own workspaces (list_workspaces), optionally straight into a folder of that workspace (list_folders).
    delete_transcriptions✓✓Transcribe and organise · can overwrite or removeDelete transcripts from the person's library, the same as deleting them on the website: each finished one goes to the website's Trash, where the person can restore it for 30 days (you cannot see or restore Trash).
    create_folder✓✓Transcribe and organiseCreate a folder in one of the person's workspaces (default: their personal workspace), the same as on the website.
    rename_folder✓✓Transcribe and organiseRename one of the person's folders (list_folders).
    delete_folder✓✓Transcribe and organise · asks you first, can overwrite or removeDelete one of the person's folders AND the transcripts in it: they go to the website's Trash, where the person can restore them for 30 days (you cannot see or restore Trash).
    create_workspace✓✓Transcribe and organiseCreate a new workspace for the person, the same as on the website.
    rename_workspace✓✓Transcribe and organiseRename one of the person's workspaces (list_workspaces).
    delete_workspace✓✓Transcribe and organise · asks you first, can overwrite or removeDelete one of the person's workspaces, the same as on the website: its finished transcripts go to Trash, where the person can restore them for 30 days (a restored one comes back in the personal workspace; you cannot see or restore Trash).
    get_meeting_participants✓✓Everything · asks you firstFor ONE recorded meeting, list who took part: the display names the meeting bot heard, and any email addresses this account already holds for that meeting (the person's calendar invite, or addresses they shared the summary with).
    send_meeting_summary✓✓Everything · asks you firstEmail the summary of ONE recorded meeting to the people the person names, from Scriptivox: each recipient gets the meeting summary and a link to the meeting's shared page, once (an address that already got it is skipped).
    get_meeting_summary✓✓Read onlyRead the summary of ONE meeting recorded by the Scriptivox meeting bot, the same summary the website shows on the meeting's Summary tab: its purpose, key takeaways, action items and next steps with who owns each and when in the meeting it came up, and the topics discussed.
    open_meeting_share✓✓EverythingShow ONE recorded meeting's share link to the person, in a box in the chat: the same link as the website's Share button, with Copy and Stop sharing, or a button to make one if the meeting has none.
    list_tags✓✓Read onlyList the tags on the signed-in person's account: the ones actually in use on transcriptions with a count of each, and separately the named-tag registry the web app maintains.
    tag_transcriptions✓✓Transcribe and organiseAdd tags to up to 10 existing transcriptions at once: the tool for "label these interviews as Q3".
    list_folders✓✓Read onlyList the folders on the signed-in person's account, with the workspace each belongs to.
    move_to_folder✓✓Transcribe and organiseFile up to 50 existing transcriptions into a folder, or pass folder_id: null to move them out of any folder.
    list_workspaces✓✓Read onlyList the signed-in person's workspaces.
    get_transcript_audio✓✓Read onlyReturn a time-limited signed URL for the source audio or video of an existing transcription, so it can be handed to another tool or streamed.
    get_transcript✓✓Read onlyRead the text of a completed transcription in the person's Scriptivox library, with timestamps and speaker names.
    chat_with_transcript✓EverythingAsk a question about an existing transcript and get an answer from the model, with the transcript as context.
    list_automations✓✓Read onlyList the automations on the signed-in person's account: saved chains of steps that run over a finished transcript (built on the website; creating, editing and deleting them is done there).
    run_automation✓✓Everything · asks you firstRun one of the signed-in person's existing automations over one completed transcription.
    get_automation_run✓✓Read onlyCheck the progress of an automation run started by run_automation: overall status, plus each step with its status, model, duration and error.
    start_meeting_bot✓✓Everything · asks you firstSend a bot to join ONE Zoom, Google Meet, Teams or Webex call and record it, producing a speaker-attributed transcript after the call ends.
    stop_meeting_bot✓✓Everything · asks you first, can overwrite or removeTell a meeting bot that is currently in a call to leave.
    cancel_scheduled_bot✓✓Everything · asks you first, can overwrite or removeCancel a meeting bot that has not joined yet, so it never joins.
    list_scheduled_meetings✓✓Read onlyList the signed-in person's meeting bots that are scheduled or currently running, plus any dispatches still queued waiting for a free bot.
    list_calendar_meetings✓✓Read onlyList the upcoming meetings on the calendars the person connected on the Scriptivox website (Google or Outlook), with whether the Scriptivox bot will join and record each one, and whether "record all meetings" is on.
    set_meeting_recording✓✓EverythingSwitch the Scriptivox bot on or off for meetings on the person's connected calendar, the same as the switch beside each meeting on the website: on, the bot joins that meeting and records it, and the transcript appears in the library afterwards (it uses the account's meeting minutes); off, it does not join.
    set_auto_record✓✓Everything · asks you firstTurn the website's "record all meetings" switch on or off for every calendar the person connected.
    upload_file✓✓Transcribe and organiseShow an upload box in the chat so the person can drop an audio or video file from their computer to transcribe it (up to 5 GB).
    open_library✓✓Read onlyShow the person's Scriptivox library in the chat (their transcripts, folders and tags) so they can browse, search and open one.
    open_subtitle_file✓✓Read onlyOpen a .srt or .vtt subtitle file in the Scriptivox subtitle editor (ChatGPT desktop opens such files with it).
    read_settings✓✓Read onlyRead the person's Scriptivox transcription defaults: default language and speaker labels.
    update_settings✓✓Transcribe and organiseChange the person's Scriptivox transcription defaults: default language and speaker labels.
    edit_subtitles✓✓Read onlyOpen a subtitle editor in the chat for a completed transcription: the person can correct the text and timing of each caption and download SRT or WebVTT.
    open_meeting_bot✓✓EverythingShow the meeting-bot card in the chat: the bots that are scheduled or recording now, and a form the person fills in to send a bot to a Zoom, Google Meet, Teams or Webex call.
    search_docs✓Read onlySearch the Scriptivox documentation and agent guides.
    get_doc✓Read onlyFetch a full Scriptivox documentation page as markdown, by its slug (for example "quickstart", "authentication", "api-reference").

    How it behaves

    • transcribe_url returns quickly. On a plan connection it returns a transcription_id at once and the transcript is read with get_transcript (the progress card in the chat says when). On API credits it waits up to 25 seconds for a signed-in app or 55 seconds for a key, then returns the id and tells the app to call transcribe_status. The job itself keeps running.
    • Stateless. No Mcp-Session-Id; every request stands alone and gets one JSON response (no SSE stream). GET and DELETE answer 405. Request bodies over 1 MB are refused.
    • Protocol revisions: 2026-07-28, 2025-11-25, 2025-06-18, 2025-03-26. For 2026-07-28 there is no handshake: server/discover lists versions and capabilities, and each request carries its version in _meta.
    • Also offered: prompts (meeting-notes, transcribe-audio, summarize-latest-meeting, make-subtitles, find-in-transcripts; Claude on a plan connection also gets transcribe-a-file, which opens the upload box), the in-chat screen as an MCP Apps resource, and MCP Events (events/list, events/subscribe) for signed-in apps.
    • https://www.scriptivox.com/mcp answers too; platform. is the address to use. Apps connected before 12 Oct 2026 at /mcp/directory keep working there.

    Checking it by hand

    bash
    curl -s https://platform.scriptivox.com/mcp \
    -H 'Content-Type: application/json' \
    -H 'MCP-Protocol-Version: 2026-07-28' \
    -H 'Authorization: Bearer sk_live_YOUR_KEY' \
    -d '{"jsonrpc":"2.0","id":1,"method":"server/discover"}'
    curl -s https://platform.scriptivox.com/mcp \
    -H 'Content-Type: application/json' \
    -H 'MCP-Protocol-Version: 2025-11-25' \
    -H 'Authorization: Bearer sk_live_YOUR_KEY' \
    -d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'

    Leave the Authorization header off and you get the 401 challenge that starts sign-in.

    The documentation server

    https://platform.scriptivox.com/mcp/docs speaks the same protocol and has exactly two tools, search_docs and get_doc, over these pages. It needs no credential, touches no account, and changes nothing. They are on the main address too, for API-credits connections and keys. The scoping is a clear contract, not a permission boundary.

    Tools in the page itself (WebMCP)

    The servers above serve an agent calling Scriptivox from outside. WebMCP is the other case: an agent driving the browser a person is already signed into. Scriptivox registers its page tools with document.modelContext.registerTool() (navigator.modelContext is accepted as a fallback):

    ToolWhat it does
    scriptivox_whoamiWhether anybody is signed in on this page, and as whom. Call it first.
    scriptivox_open_signupOpens the signup form. It does not create an account.
    scriptivox_get_plan_statusThe signed-in person's plan and entitlements.
    scriptivox_search_transcriptsFinds the person's transcripts by name. Read only.
    scriptivox_open_transcriptOpens one of them in the page.
    scriptivox_transcribe_linkFills in a link to transcribe; the person confirms on the page.
    scriptivox_start_plan_checkoutReturns a Stripe Checkout URL. Charges nothing.
    scriptivox_create_api_keyAsks the person, on the page, to create an API key.

    Nothing that spends, sends or creates a key happens without the person clicking on the page. WebMCP is a W3C draft and absent in most browsers today; the page registers what it can and does nothing when the API is missing.

    The old local package

    @scriptivox/mcp-server (npm) and its Docker image are retired in favour of the hosted address above, which has every tool and nothing to install. Existing installs keep working with an API key; new tools arrive only on the hosted server.

    Versioning

    The hosted server follows the MCP spec's revisions, listed above; a tool renamed or removed is announced in the changelog first. The 1.0.x names transcription_url and transcription_status still work but are no longer listed.