Every call your company recorded, transcribed once and readable from Claude. Search a transcript for a phrase, or ask across the archive from chat — then click the timestamp to hear it said out loud.
MCP connector · answers cite clickable timestamps · never used to train a model · paid plans only
Somebody asks why the vendor decision went the way it did. Three people were on the call, two have moved teams, and the recording is in a folder nobody has opened since. The company already paid for that conversation once; the cost of not being able to find it again is paying for it twice — re-litigating a settled decision with worse information.
Which layer you want depends on whether you know the phrase, know the call, or know neither.
Type it and get every line that contains it, each with a timestamp, so you judge relevance before you jump. Case and accents are folded, so codigo finds código — which starts to matter the moment your archive is not all in one language.
Open a transcript and ask a question of it. The answer is grounded in that transcript only, and every citation is a link: click it and the transcript jumps to the line while the audio plays from it. You check the passage rather than trusting the paraphrase.
“What did we commit to on the integration timeline, and who owned it?”This is the part no meeting-notes tool does. Connect Whipscribe to Claude once and your archive becomes something you interrogate in chat: Claude lists your transcripts, pulls the ones that look relevant, searches inside them and answers over their text — then names the file and the timestamp so you can go and listen.
Point it at a library folder to scope the question. Ask it to rename files so File 12 becomes Vendor review — March 14, and next quarter's question is easier than this one.
No new interface to learn — you are asking in the tool you already have open.
It was settled in Leadership weekly — March 14. The integration was cut rather than delayed, because the security review would not close before the launch date. The commitment is stated at ◷ 00:34:12 and restated as an action item at ◷ 00:41:05. An earlier call on Feb 28 ◷ 00:12:40 is where the security concern first comes up.
Setup is two fields: in Claude open Settings → Connectors → Add custom connector, name it Whipscribe, URL https://whipscribe.com/mcp. Step-by-step with screenshots on the connector page. The same connector works in Claude Desktop, Claude Code, Cursor and any other MCP client.
The tools your archive exposes to chat, in the words of the server that serves them.
| Tool | What it does |
|---|---|
list_my_transcripts | Every transcript in the account, including ones uploaded through the web UI — not just what chat submitted. |
get_transcript | Reads one transcript back as txt, json, srt, vtt or docx. |
clips_search_transcript | Searches the sentences of one transcript and returns matches with start and end times plus the speaker. |
rename_transcript | Gives a file a human title that shows up immediately in your file list. |
library_list_folders · library_add_item | Organizes the archive into folders, so a question can be scoped to one team or quarter. |
library_share_folder | Mints a read-only link to a folder — 30 days by default, up to 365. |
transcribe_urls_batch | Adds up to 20 new recordings by URL in one call, run concurrently. |
This is not a single vector index you type a query into. Whipscribe holds the transcripts and the literal in-transcript search; the cross-archive reasoning happens in Claude, over the transcripts it pulls. In practice that is why folders and real file names matter — a well-scoped folder gives a faster, better-grounded answer than a thousand files called recording_final_2.mp4.
The connector reaches exactly one thing: your Whipscribe account. Not Slack, not email, not Drive. And it is scoped to the account bound to your key — a question can only be answered from meetings that were actually recorded and transcribed.
The pattern is the same each time: a decision, or a customer's exact words, months later, wanted by someone who was not in the room.
What the customer actually objected to, in their words, before the renewal conversation — rather than what the CRM note says they objected to.
The decision trail. When a commitment was made, what the stated reason was, and which constraint had changed by the following month.
Ask across thirty interviews for the pain point that keeps recurring, then pull the three cleanest quotes with their timestamps.
Find the calls where a specific promise was made, and read the exchange rather than the ticket summary of the exchange.
A shared folder of the calls that explain how the product got this way — read at reading speed, not listened to at talking speed.
Compare what candidates actually said across a panel's recordings, against the transcript rather than against memory.
You pay once per minute of audio to make it readable, and the minutes do not expire. There is no free tier.
| Pack | Price | Minutes | Roughly |
|---|---|---|---|
| Unlock | $2 | 60 | one hour — one call, to see the shape of it |
| Starter | $4 | 300 | 5 hours |
| Pro | $12 | 2,000 | 33 hours |
| Team | $24 | 5,000 | 83 hours |
104 weekly meetings averaging 55 minutes is 5,720 minutes, about 95 audio-hours. One Team pack ($24 for 5,000 minutes) plus one Pro pack ($12 for 2,000 minutes) is $36 for 7,000 minutes — it covers the whole two years with 1,280 minutes still in the account. That is roughly 35 cents per meeting to have a decision record you can query.
Keeping it current: 40 sales calls a month averaging 30 minutes is 1,200 minutes a month. One Team pack — $24 — covers about four months of that.
Search, questioning and every export format are included on every pack. Full pricing →
Transcribe one recording, connect Claude, and ask it the question that made you open this page. If the answer comes back with a timestamp that holds up, the rest of the archive is the same job repeated.
Transcribe a call →Not in the sense of a single vector index you type into. Whipscribe holds the transcripts; the cross-archive question-answering happens in Claude through our connector, which lists your transcripts, pulls the relevant ones and answers over their text. Inside a single transcript the browser gives you literal keyword search with a timestamp on every hit. We describe it this way on purpose — you should know where the reasoning happens before you build a process on it.
In Claude, open Settings → Connectors → Add custom connector and paste the name Whipscribe with the URL https://whipscribe.com/mcp. Every chat after that can list, read, search and organize your transcripts. Screenshots on the connector page.
No. The connector reaches exactly one thing: the transcripts and library items in your Whipscribe account. If a meeting was never recorded and transcribed, nothing here knows it happened.
Two ways. Scripted uploads carrying the X-User-Email header all land in one shared account, so everyone reads the same library — see bulk meeting recordings for that script. For people outside the account, mint a read-only share link on a library folder: 30 days by default, up to 365.
No. Transcription runs on open-source Whisper models on our own infrastructure; your audio is never sent to a third-party transcription service and is never used to train a model. On a paid plan uploaded audio is kept 365 days after the job completes, and transcripts stay until you delete them.
No — no bot, no participant, no calendar access. You record the way you already record and hand over the file. The Chrome extension can capture a browser tab's audio into your account, which is still a recording rather than an attendee.
104 meetings averaging 55 minutes is 5,720 minutes, about 95 audio-hours. One Team pack ($24 / 5,000 minutes) plus one Pro pack ($12 / 2,000 minutes) is $36 for 7,000 minutes — roughly 35 cents per meeting, with minutes left over that do not expire.
The script that gets 300 existing calls into the archive in one pass.
Connector setup, screenshot by screenshot.
Summary, decisions and action items for a single call.
Folders, items and share links for the archive itself.
Where the local recording lands and what to do with it.
Submit, poll, fetch, retention — for your own indexing.
The wider vertical this page sits in.
$2 / 60 min · $8 / 1,000 min · $12 / 2,000 min · $24 / 5,000 min.