neuroflash MCP - Let your AI agent ask 1M real people before it writes

Hey Product Hunt

I'm Donald, I build the API and MCP at neuroflash. Quick story on why this exists.

For six years neuroflash was an app that helped marketers write content. The quiet part was the engine underneath: a panel of more than a million real people we have surveyed and validated across 80+ academic studies. Internally we used it to predict how content would land before we published it.

For the last year I kept hitting the same wall. Every agent I built, in Cursor, in Claude, in our own stack, was writing for an audience it had never met. The usual fix is to prompt a persona. But a prompted persona is just a language model imagining a person, and against real human response that lands around 55 percent. A coin flip, right when real budget is on the line.

So we opened the panel.

neuroflash MCP lets any agent ask real people before it writes a word. Send a headline, a slogan or an ad concept to a real audience segment and get back a score, the reasoning, and how many people answered. And it is not only audience. Brand voice, content and images live behind the same connector. Four things your agent could not do on its own, through one URL.

What is live today:

A hosted MCP server that works with Claude Desktop, Cursor and Windsurf. Paste one URL, sign in once, start asking. Nothing to install.

A REST API, 88 endpoints across seven domains, behind a single key.

A free developer tier, 5,000 credits a day, no credit card.

What it is not: another wrapper around an LLM. The model still writes. We just give it real people to write for.

I am here all day and I want the hard questions. Ask me about the data, the accuracy, the auth, anything.

This is the part I have been waiting to talk about.

Donald

Gordian here, 👋 Head of Product at neuroflash.
Donald and the team built this — I want to add the part I find more interesting than the launch itself.

Look at what an agent can actually call today. Databases. Search. Code. Payments. Every one of those is data about systems. Almost nothing in the stack lets an agent get data about people — and yet most of what we ask agents to produce is aimed at people.

That's a gap we sat on for six years without noticing it was a gap. We had a validated panel and we used it inside our own app, as a feature. MCP is what made it obvious the panel was never a feature. It was infrastructure with a UI bolted on top.

Fair warning: we're new at being a developer product. Tell us where it's rough.

Dev on the neuroflash team here 👋🏻

The tool I keep reaching for is `chat_with_twin_group`: one call to a demographic group and you get individual answers with reasons back.
I wired it into a `pre-publish` check that scores copy against the target segment before anything ships.
Setup was genuinely zero config, added the connector in Claude Code, signed-in once, done.

Happy to answer anything about the API shape or the structured output ✌🏻🤓

hey its a great idea can you tell more about the where did you train the real human data