Imagine you were on a call with a client last week, and they promised “a first draft by next Wednesday.” You recorded it with Speechyou at the time, and the transcript sits in its backend. By Monday you can’t remember whether it was Wednesday or Thursday, so you open Claude and ask. It doesn’t ask you to upload a file; instead, it goes to a place called Speechyou and searches your voice notes, finds that transcript, quotes the exact words, and attaches a link to the original audio.
This is one concrete action Speechyou performs: connecting voice notes and meeting transcripts into Claude. The connection uses OAuth 2.1, not an API key; it can only access notes you own or that someone has shared with you. To use this custom connector, Claude requires a paid plan on its side; Speechyou itself can be started for free. [1]
Image: Source product site[2], accessed 2026-11-11; original interface retained.
Speechyou itself is a transcription tool, and this connector is another outward-facing entry point. You can record directly in the browser; meeting mode captures both microphone and system audio, covering Zoom, Teams, and Google Meet. The recorded audio is sent to Whisper and the proprietary MultiLingual Pro model for transcription; the site calls the transcription accurate but does not explain how the two models divide the work, nor does it provide figures that can be checked. After transcription, you can use a chat-style AI to get summaries, action items, and key points, and you can also ask questions directly. [2]
This connector’s permissions are tightly scoped. When a transcription product hands an entry point to another AI, the permissions could have been broader: letting it write back, organize, or even delete old recordings. Speechyou did not do that. The site places the read-only boundary under its security explanation, but does not say why it does not offer broader permissions such as write-back, organizing, or deletion; what is certain is that this boundary lets Claude only find and read things, not edit meeting notes for you or delete recordings for you.
Pricing information is listed under two names across two pages. The product page calls it Solo: $15/month, $67/year when paid annually (about $5.58/month, billed annually), and a 3-day trial covers all features in this tier. The pricing page calls it Unlimited: $15/month, $100/year when paid annually (about $8.33/month, billed annually), and a 3-day trial covers all Unlimited features. Monthly billing is $15 on both pages, but the annual prices are different and the names are different; the site does not say whether they are the same plan. The product page says you can cancel at any time; after the trial ends, you need to choose monthly or annual billing, but neither page states when charges occur or the refund terms. Subscription payments are processed through Stripe and Apple Pay, and transcription uses third-party AI services; both points appear in the privacy policy updated on January 23, 2026, not on the pricing page. [2] · [3] · [4]
On TrustMRR, founder Cornelius states in a comment that the actual monthly recurring revenue (MRR) is 1800$, average selling price (ASP) 15$, and lifetime value (LTV) 80$; in the same comment, he says all traffic comes from AI and Google SEO, with no paid media. A record updated on November 14, 2026 on the same platform shows MRR $1,351, 104 active subscriptions, and $8,341 in all-time revenue, and says the revenue was verified via a Stripe API key. The two figures use different bases—one is the founder’s self-report, the other is the platform record—and I do not merge them into one. [5]
The same record also lists: the product is for sale, priced at $35,000, labeled 1.4x revenue, with 18 interested buyers; founder Cornelius P, team of 1, Bootstrapped. Costs for transcription, storage, payments, support, and staff are not disclosed, so the MRR of $1,351 and 104 subscriptions do not show profitability; they are just two revenue-side numbers. The material does not disclose why he chose to sell at this point, nor does it disclose funding history or the sale process.
A one-person team puts revenue, unit price, lifetime value, acquisition channel, and sale price on the same public page. On the product side, the connector keeps permissions tight; on the business side, he lays out the numbers and puts up a price tag. All that can be seen are these choices. The site also lists an affiliate program: approved partners can earn a 30% commission on qualifying purchases attributed to their referrals; the material does not say how much revenue it has brought in. [6]
Turning meetings into text is not something only Speechyou does. Fathom’s free tier also covers unlimited recording and transcription and instant AI call summaries, with capture available via bot-free (beta) or bot; the two overlap on “meeting transcription,” but Fathom’s feature table also lists Claude and ChatGPT integrations, a Public API, and MCP; these are features listed on the page, not same-condition tests. The more direct difference is the entry condition: Fathom has a free tier that is free long-term, while Speechyou requires choosing monthly ($15/month) or annual after a 3-day trial, and its own two pages’ annual numbers still do not match. Which has more features can only be judged by what each side’s pages list. [7]
Transcription itself is no longer novel; what is hard is where those words get used after transcription. Speechyou’s MCP connector does this well: you ask Claude, and it goes through your notes, finds it, quotes it, and attaches the original audio. The action is small, but it is concrete.