Free account: 350 credits for 30 days, no payment details

99takes in Claude Code:
test copy before you spend budget on it.

You connect 99takes to Claude Code with one command. From then on, Claude tests headlines, ads and entire landing pages against personas from your audience, right in the same conversation. You see what holds them back and which version they prefer.

The problem

Claude can write anything. It just can't know whether it will land with your buyers.

It doesn't know your buyers

It knows language, patterns and plenty of landing pages. But it only knows what makes your buyers hesitate over this particular decision once someone has researched it.

What you learned leaves with the chat

Which version already won for you doesn't carry over to the next chat on its own. With 99takes, your audience, runs and reports stay in your account, and a new conversation picks up from there.

A model grading itself tends to agree with itself

Ask "Is this copy good?" and you get a pleasant, plausible answer, even about copy the model wrote a moment ago. Whether your buyers would agree doesn't come into it.

How it works

One command, then it's part of how you work.

Connect once, and from then on the tool lives where the copy gets written.

Set up once
claude mcp add --transport http 99takes https://api.99takes.com/functions/v1/mcp

On the first call you sign in through your browser. With a key from your account, no sign-in is needed:

claude mcp add --transport http 99takes https://api.99takes.com/functions/v1/mcp --header "x-api-key: YOUR_KEY"

Then ask Claude: "Are you connected to 99takes? How many credits do I have?"

Then, in the conversation
> Write us three headlines for the HRIS landing page.

  ⚙ write_challenger
  1. The employee record that keeps itself up to date.
  2. Leave, references, contracts: all in one record.
  3. Less upkeep on employee records, every week.

> Test the first one against our current headline.

  ⚙ simulate_copy
  Version 1 LOSES to the current headline.
  Score 5.1 vs 6.0, 14 of 36 sim runs for version 1.

  Objections, lowest score first:
  • Head of IT, 120 employees (3.8/10): Doesn't say what
    it replaces. I can't tell if this sits next to our
    HR system or takes it over.
    Source: review · example reference
  • Head of People, 400 employees (4.6/10): "Keeps itself
    up to date" reads like a promise that lands on me in
    the next audit.

Example with made-up data.

You create the key in your account at run.99takes.com. The setup page walks through every step, including claude.ai.

What you can ask

What 99takes answers in Claude Code

You tell Claude what you want to know. 99takes tells you the price in credits before every run, and nothing is charged until you say yes.

Analyze a landing page

Copy, structure and reading flow, how it comes across on desktop and mobile, a competitor check and an action plan.

price shown before you start

Test a new version against your current page

99takes writes a challenger version. You see it first, then it runs against your page.

price shown before you start

Compare two pieces of copy

Headlines, ads, sections or subject lines, with a verdict and objections per persona.

about 73 credits, 1:1 about 19

Have versions written

99takes writes from your audience research. Three headlines cost about 6 credits.

4 to 18 credits

Check core claims against competitors

Which of your claims are open, contested or taken, with a verbatim quote.

3 to 7 credits

Compare above-the-fold drafts

Two to four drafts of what visitors see before scrolling, and which one the personas would click.

up to 12 credits for two drafts, up to 24 for four

Research your audience

10 to 14 personas covering a wide range, backed by real customer voices where they exist.

15 credits

Share a report

Send the result to your team or client as a link, and revoke it at any time.

You'll find questions to copy word for word under example questions.

Why it belongs in Claude

Checking is half the job. Rewriting is the other half.

People read a report and file it away. In Claude Code, the objections land where the copy gets written: Claude reads them, has a new version written and checks again, all in the same conversation.

Step 1

Write

You bring your draft, or 99takes writes challenger versions from your audience research.

Step 2

Collect the pushback

10 to 14 personas read it, each also with the other personas’ jobs in mind, and tell you in one sentence where they get stuck.

Step 3

Check the rewrite

The new version runs against the old one. You see whether the personas prefer it or whether it's just different.

Where objections come from

Every objection backed by evidence shows its source.

A persona here isn't an invented role with a name and a stock photo. It draws on public customer voices, plus your own material if you add it. An objection based on that evidence says where it came from.

Where an objection has no evidence behind it, none is shown. Better no source than one that doesn't exist.

What a citation looks like (reconstructed, not a real quote)

"To this day I don't know whether that thing replaces our HR system or sits next to it. That's the reason we didn't take it further."

review · example reference

Built for

Teams who want to test copy where they write it.

Growth and dev teams

You build landing pages with Claude Code and test the copy in the same workflow, before it goes live.

Freelancers

For a tool you need twice a month, a credit pack with no subscription does the job. Purchased credits are valid for 36 months.

Agencies

A second opinion before client sign-off, and a report link that shows why you chose this version. On Pro and above, reports can be limited to invited readers.

What it isn't good for

A simulation doesn't replace a measurement.

If your page has traffic and conversions, your own numbers are the better source. Test in the market then. The value comes before that, in the window when nothing can be measured yet because nothing is running.

It isn't an oracle either. It tells you which of two versions these personas prefer and where they get stuck. Whether your offer itself is right isn't something it can answer.

Access

Create an account and test it yourself

The account is free and starts with 350 credits, valid for 30 days, no payment details needed.

In claude.ai, you add 99takes as a connector, no key required. In Claude Code, you run one command.

For every step, a test prompt, and troubleshooting, see the setup page. For what a run costs after that, see the pricing page.

FAQ

What people want to know first

What does 99takes do in Claude Code?

It gives Claude tools to test copy and pages against personas from your audience. You give it a piece of copy or a URL, the personas read it, and you get their objections back. With two versions, you also see which one they prefer. If you like, 99takes writes new versions for you as well.

How do I set up 99takes in Claude Code?

Create an account at run.99takes.com and run one command in your terminal: claude mcp add --transport http 99takes https://api.99takes.com/functions/v1/mcp. On the first call, you sign in through your browser. Or add a key from your account to the command instead. To check that it worked, ask Claude: "Are you connected to 99takes? How many credits do I have?"

Why not just ask Claude whether the copy is good?

A model that just wrote a piece of copy judges it with the same knowledge it used to write it. It doesn't know what your buyers actually said. 99takes tests against a researched audience that stays the same between runs, so results can be compared.

Why a comparison instead of a score?

A score for a single piece of copy has no yardstick: it can't tell you whether a 6.4 is good or bad. Two versions side by side pose a question that has an answer: which one do the personas prefer? If you don't have a second version, 99takes puts one up against yours.

Where do the personas come from?

From research into your market: public customer voices such as reviews and forum posts, plus your own material if you add it. An audience has 10 to 14 personas. Statements backed by evidence carry their source, and any objection that relies on one passes that source on to you. Where there is no evidence, no source is shown.

Does it work in claude.ai too?

Yes. In claude.ai you add 99takes as a connector, without a key, and approve the connection at run.99takes.com. Account, credits and audiences are the same in both.

Does this replace A/B testing?

No, it comes before it. An A/B test needs traffic, which new copy doesn't have yet. By the time a result comes in, you've already spent budget on the copy you wanted to check. Once you have traffic and conversions, your own data is the better source.

What does it cost?

You pay in credits. The account starts free with 350 credits, valid for 30 days, no payment details. Comparing two versions with 12 personas, each also reading with the other personas’ jobs in mind, costs about 73 credits, or about 19 1:1, with each persona reading with only its own job in mind. After that, you buy credits from €17.85 including VAT, with no subscription. See all prices at https://www.99takes.com/pricing.

For everything you can check in a conversation, from two versions to a whole page, see buyer simulation. How 99takes writes new versions: challenger versions. Why we check copy before the traffic arrives is explained on the A/B test alternative page.