Skip to content

I stopped typing.
This is the rail I built after that.

Four Skills to install, a one-week plan, and a one-pager on MCP. No form, no sign-up. 5,410 dictations and 86 meetings behind it.

A voice turning into a rail: from speech to a terminal, meetings and documents

Voice-rail check, 30 seconds

Five yes-or-no questions. Answers stay in your browser; only the score is counted.

Do you dictate the long prompts instead of typing them?Do you dictate the long prompts instead of typing them?
Does every client meeting sit on a calendar invite?Does every client meeting sit on a calendar invite?
Do summaries land where you actually work?Do summaries land where you actually work?
Do numbers reach a document only after a look at the full transcript?Do numbers reach a document only after a look at the full transcript?
Does your agent read the meetings itself (MCP)?Does your agent read the meetings itself (MCP)?

Four Skills. One input, one output.

A Skill is a fixed instruction with one input and one output. Install once, type its name, give it the input, get a fixed-shape result. Nothing is ever sent on its own.

Post-call

/post-call

A transcript goes in. A follow-up message and a one-page scope come out. Nothing is sent.

Client context

/client-context

A name goes in. One board with everything known, conflicts between sources included.

Prep call

/prep-call

A name and a date go in. A seven-section cheatsheet in the right register comes out.

Sync call state

/sync-call-state

A reviewed recap goes in. One diff comes out and waits for your yes.

Open source, MIT. Two lines in Claude Code; in any other agent, copy the folder.

/plugin marketplace add yonatangross/rail-kit
/plugin install rail-kit@yonyon
cp -r rail-kit/skills/* ~/.claude/skills/

Source, tests and examples: github.com/yonatangross/rail-kit (version 1.0.0).

No Wispr Flow yet?

It is the tool the rail runs on: dictation in every app, a Notetaker for meetings, and an MCP that lets your agent read them. Signing up through this link gives you a free month of Pro (and one to me).

Start with Wispr Flow, a free month of ProYonatan's referral link. The free month is applied automatically.

official docs docs.wisprflow.ai

Your meetings, readable by any agent

The Skills work on anything you paste. They work far better when the agent reads the meetings itself, which is what the Wispr Flow MCP does. Mac, Notetaker beta.

Calendar meetings flowing along a rail into an agent that reads them
  1. 01On Mac: Wispr Flow settings, then MCP (after Connectors).
  2. 02Claude or ChatGPT: one click on Connect, approve in the browser, done.
  3. 03Any other tool (Cursor, VS Code, Claude Code): the URL under All other apps, as a new MCP server.
  4. 04What the agent sees: meeting summaries and transcripts, notes, tasks, calendar. Read-only.

One week, without building anything

  • Day 1: install and the shortcut. Ten minutes of dictation instead of typing, only for the long prompts.
  • Day 2: personal dictionary. Internal terms, product names, the words that keep coming up.
  • Day 3: meetings. Turn the Notetaker on, and put every client meeting on a calendar invite.
  • Day 4: one place where summaries land, which is the place where you actually work.
  • Day 5: the rule: numbers move only with a transcript. Written somewhere visible.
Five lit stepping stones leading to a dock with a microphone

Want the rail running on its own for you?

Summaries that land in a knowledge base by themselves, dictation that triggers an agent, the Skills tuned to your business. 20 minutes is enough to know if it fits.

Book a free 15-minute callNo sales deck. You leave with a map, even if we never work together.