Mimiq MCP server: user testing from your coding agent

Coding agents ship pages and sign-up flows in minutes, and nobody outside the team sees them until real visitors do. A test inside the agent's loop puts skeptical simulated people in front of the change while it is still cheap to fix, and the agent can read their objections and act on them before you review the diff.

• Eight tools: test a page, run a flow, test copy, text or a UI component, compare two pages or two pieces of copy, and ask a question.
• Works with Claude Code, Codex, Cursor, VS Code, Windsurf, and Claude Desktop.
• One free test without a key; a free account adds 100 credits.
• With a key, tests and comparisons are saved to your account, with a link to the full report.

In Claude Code, run this once in your project. Leave out the header to use the free test, or set MIMIQ_API_KEY to a key from your Mimiq settings.

claude mcp add --transport http mimiq https://mcp.mimiqai.com/mcp --header "Authorization: Bearer $MIMIQ_API_KEY"

Clients that read an MCP config file take the same address. In Cursor, put this in .cursor/mcp.json, and keep a file with a key out of your repository. The README at github.com/victorgulchenko/mimiq-mcp has the setup for Codex, VS Code, Windsurf, and Claude Desktop.

{
  "mcpServers": {
    "mimiq": {
      "url": "https://mcp.mimiqai.com/mcp",
      "headers": { "Authorization": "Bearer mq_sk_..." }
    }
  }
}

Ask in plain words: test my landing page on startup founders, compare these two headlines on freelance designers, or compare the preview deployment with production. The agent picks the tool.

A flow sends each person through a task in their own real browser, such as signing up, for up to 15 steps by default, and returns each person's steps, where they stopped, and why. It uses 30 credits a person; without a key, a free flow has up to 3 people, within a daily limit.

A comparison shows both versions to the same simulated people and returns Mimiq's call: clear, leaning, or too close to call, with how many people moved each way.

Everyone in a Mimiq result is simulated. On 1,000 real headline A/B tests held out from its development, Mimiq's forecast picked the headline more readers clicked 76% of the time; the best rule of thumb, picking the headline written first, got 61%. Mimiq's forecast has held up on headlines. In pre-registered tests on emails, text messages and ads, it was no better than chance at ranking small wording changes, and on real web page tests it did not beat the simplest guess. So treat a call on copy, an email, or a page as a fast first read, not a measured result.

A flow shows where people like your visitors can get stuck, not how many real visitors will finish. It is a usability check, not a conversion forecast.

Mimiq opens pages in a real browser on its side, so the address has to be public. For a local dev server, open a temporary tunnel, such as a free cloudflared tunnel, and pass its address. The page is public while the tunnel runs, so your agent should tell you before it opens one.

To get a key, create a free account (it adds 100 credits), open Settings, find "Use Mimiq from your coding agent", and choose "Create a key". Send it as "Authorization: Bearer mq_sk_..." on every request. Packs start at $29 for 500 credits, and credits never expire.