ScriptGrain makes Claude better at being you

By Jack Stovell X · Instagram · LinkedIn · published 2026-09-20 · updated 2026-09-30

Claude can write for you. It cannot measure you. ScriptGrain measures your voice as 45 attributes from writing you have already published, connects to Claude, and scores every draft against the measurement, so the assistant you already use writes as you rather than as its average.

The next step is the loop itself. We lay it out in a draft-score-publish workflow: draft in Claude, score against your profile, polish anything under 0.85, save the edit, publish. It all runs over MCP, and the scoring is free every time, so checking your own work never burns a credit.

Start with the prompt builder below: paste your writing and get a system prompt written in measured numbers, ready for custom instructions or a project. The connector does the same with the full profile.

How to connect

Works with Claude (claude.ai, Claude Desktop and Claude Code).

  1. Measure your voice. Create a free ScriptGrain account and paste three or more pieces you wrote (3,000 words works best). The profile usually takes under a minute and costs nothing.
  2. Connect. In claude.ai open Settings, then Connectors, then Add custom connector, and give it `https://mcp.scriptgrain.com/mcp`. Claude opens ScriptGrain's consent page; sign in and the connector is live. In Claude Code run `claude mcp add scriptgrain https://mcp.scriptgrain.com/mcp` and sign in the same way.
  3. Write, then score. Claude can build the profile too: give it three or more pieces of your writing and ask it to call `create_profile`, then `get_profile` until the status is complete. From then on every draft can be generated in the voice, scored with `check_voice_match`, and polished toward 0.9.

What changes

Without a profile, Claude writes from a description: whatever you can say about your own style. Most of a voice is what you cannot say: contraction rate, clause ordering, how often you open with a question, the words you reach for without noticing. With ScriptGrain connected, Claude writes from the measured profile and every draft comes back with a voice-match score and the exact features that moved. On-voice drafts score 0.85 to 0.95; a clearly different voice scores under 0.45. Paste any two pieces into the free off-voice check to see the same measurement with no account.

What stays free

The profile, the connector, scoring any text against your profile, and comparing pieces with no profile at all are free on every plan. Generating drafts in your voice, polishing toward a target and the monthly voice monitor are paid.

What Claude reads from the profile

Claude reads numbers. You get the measurement itself, with no paragraph of adjectives about your style: 45 attributes across eight layers, from average sentence length and its variance to contraction rate, comma density and how you open a paragraph.

Take a description. You write "conversational, a bit dry, short sentences" and Claude does its best with that. Fair enough. But it has no way to know your contraction rate, or that you put conclusions first and build toward them, or that you reach for "but" over "however" without noticing.

The profile is read live through the MCP server each time, so it's never a stale paste. `get_profile` returns the measurement. `check_voice_match` scores a draft against it, free on every plan. `generate_content` writes a draft with the profile as a constraint, at 1 credit per draft. `polish` revises toward a target. `save_edit` is free and feeds learning.

So Claude stops guessing at your voice and starts writing to a spec. The spec also works as a yardstick, which a description never could.

A drafting session, step by step

Say you've got a newsletter to write and Claude is connected. You ask for a draft on a topic of your choosing, in your voice.

  1. Claude calls `get_profile` and reads the measurement.
  2. It generates the draft with those numbers as constraints.
  3. It runs `check_voice_match` and reports the score, say 0.81, plus the features that moved.
  4. You read the deltas. "Average sentence length: draft 24, profile 16" tells you exactly what to fix.
  5. You ask Claude to polish toward 0.9, then edit whatever still jars.
  6. You save your edit, so the app learns from it.

That last step matters more than it looks. Your hand-edits are the cleanest signal of what the profile still misses.

On-voice drafts typically land between 0.85 and 0.95. A first pass sitting a little under that isn't a failure; it's the point of scoring. You get a number and a reason instead of a vague sense that something's off.

When a draft scores off voice

A low score is information. The useful part isn't the headline figure; it's the per-feature deltas, each showing the draft's value against your profile's value.

Partial matches sit between 0.5 and 0.8. A clearly different voice lands under 0.45. So a 0.6 means you're in the right neighbourhood with a few features wandering.

Read the deltas before you touch the draft. If hedging is the outlier, your profile says you state things plainly and Claude softened every claim. If sentence length has drifted, it smoothed your rhythm into something tidier than you'd ever write. Those are different fixes, and a single score would never tell you which one you need.

Then tell Claude what to correct, by feature. "Cut the hedging and break up anything over 25 words" beats "make it sound more like me" every time. Run `polish` toward 0.9, rescore, and stop when the deltas are small enough that your own edit finishes the job.

One honest caveat: a bad profile scores confidently. If the source writing was thin or mixed, the numbers reflect that. Three or more pieces and about 3,000 words is the recommendation for a reason.

Projects, Styles and the profile together

Claude's own tools are good at what they do. Projects hold documents and instructions per project (Free plans get up to 5), which makes them excellent for context: your brief, your audience, your reference material. Styles are built from a sample or a description, a quick way to nudge tone.

Neither scores output. That's the gap, and it's a narrow one. A Project tells Claude what you're writing about, and a Style suggests how it should sound. The profile measures how you actually sound, then checks the result.

So use all three. Keep the Project for background and instructions. Keep a Style if you like the shortcut. Connect the profile for the measurement and the score, since it's the only one of the three that can tell you a draft has drifted.

In practice the division is simple: context goes in the Project, a rough tone goes in a Style, and the standard of "sounds like me" is held by the profile. Each does one job. Ask the Project to do the profile's job and you're back to describing yourself, which is where most of a voice gets lost in the first place.

Questions

Do I need a paid plan to connect ScriptGrain to Claude?

No. The connector, the profile and voice-match scoring are on the Free plan. Paid plans add generation credits, polishing and the monthly voice monitor.

Does Claude send my conversations to ScriptGrain?

No. Claude calls ScriptGrain only when it uses a tool (score this text, generate this brief), and ScriptGrain receives only what that call contains. Your API key or sign-in token never appears in the conversation.

What is the voice-match score measuring?

Stylistic alignment: sentence rhythm, punctuation habits, pronoun mix, formality, argument structure and the rest of the measured attributes, compared feature by feature against your profile. It says whether the draft sounds like you, not whether it is any good.

Related