FluentaPremier
Score idea

Tutorial · 14 min

Run an X-Ray from inside Cursor (MCP).

Wire Fluenta into Cursor's agent. Free instant previews mid-conversation; full paid X-Rays when an idea earns the credits. No leaving the editor.

What you’ll build

By the end of this tutorial, your Cursor agent will be able to validate any product idea you talk about without leaving the editor. Two tools become available:

  • fluenta_idea_x-ray_sandbox — free instant preview. The agent uses it as a default reconnaissance pass.
  • fluenta_idea_x-ray — the full paid run, async (~40 min), 2,000 credits, but returns the structured 25-integration report with LRS, signal breakdown, and explicit verdict.

Total time: ~14 minutes including the install. The actual paid X-Ray takes another 20 minutes async — but you don’t have to wait at the keyboard.

Prerequisites

  • Cursor with Remote MCP support (any 2026 release).
  • A Fluenta API key with the read_write scope. The free tier ships with 2,000 credits — exactly one paid X-Ray, on the house. Create one.

Step 1 — Install the MCP server

The fastest path: click the “Install in Cursor” button on the docs page. Cursor will prompt you to confirm the install. Done.

If you’d rather paste the config manually, drop this into ~/.cursor/mcp.json (create the file if it doesn’t exist) and replace YOUR_FLUENTA_API_KEY:

~/.cursor/mcp.json
{
  "mcpServers": {
    "fluenta": {
      "url": "https://dev.fluenta.space/backend/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_FLUENTA_API_KEY"
      }
    }
  }
}

Restart Cursor after either method. Open a new chat and ask “what tools do you have?” — you should see fluenta_idea_x-ray_sandbox and fluenta_idea_x-ray in the list.

Step 2 — Free reconnaissance pass

Start with the sandbox tool every time. It costs nothing, runs in seconds, and gives the agent a verdict it can cite back. The shape of a typical exchange:

conversation: sandbox call
> User: I'm thinking of building an AI chrome extension that auto-summarises every Notion page you open.

> Cursor (calling fluenta_idea_x-ray_sandbox):
  {
    "title": "AI auto-summary chrome extension for Notion",
    "description": "Browser extension that detects Notion page loads and shows a 3-bullet AI summary in a sidebar. Free tier 10 pages/day, paid $5/mo unlimited."
  }

  → type: xray_sandbox_preview
  → verdict: "promising"
  → demand_signal: "high — 22k monthly searches for 'notion ai summary'"
  → competition: "saturated — 4 funded competitors, 2 free open-source"
  → top_risk: "Notion's own AI feature ships natively for $10/mo —
    differentiation needs a clear angle (speed? export formats?)"

The sandbox preview returns four signals: a verdict bucket (prime / promising / experimental / weak), demand signal, competitive density, and the single most likely failure mode. It does not run the full 25-integration pipeline.

Why this matters: sandbox previews give you a 90-second “is this even worth $19 of credits to investigate further?” without spending a single credit. Use it liberally.

Step 3 — Submit the full X-Ray

When the sandbox lights up green and you want the full report — LRS score, six signal dimensions, closest competitors, monetization benchmarks, explicit recommendation — ask Cursor to run the paid tool.

conversation: submit paid run
> User: OK run the full X-Ray on this. I'll spend 2,000 credits.

> Cursor (calling fluenta_idea_x-ray with read_write key):
  → type: xray_submitted
  → analysis_id: "an_01HXY4..."
  → status: "queued"
  → credits_used: 2000

> Cursor: Submitted. Run takes ~20 minutes. I'll mention it
  whenever you ask "how's the run going?" — or you can also
  read it any time at:
  https://fluenta.space/app/x-ray/an_01HXY4...

The paid tool returns immediately with type: xray_submitted and an analysis_id. The actual computation happens in the background — expect ~20 minutes wall-clock.

Heads-up: 2,000 credits are debited at submission, not at completion. If your free tier balance is below 2,000 you’ll get a 402 instead of a 201. Top up before retrying.

Step 4 — Read the result

The same fluenta_idea_x-ray tool, called with { analysis_id } instead of a fresh idea, returns the completed structured result:

conversation: poll for result
> User (40 min later): how's the X-Ray going?

> Cursor (calling fluenta_idea_x-ray with { analysis_id: "an_01HXY4..." }):
  → type: xray_result
  → status: "complete"
  → lrs: 78
  → verdict: "promising"
  → demand: { score: 84, evidence: [...] }
  → competition: { score: 62, top_competitors: [...] }
  → monetization: { benchmarks: [...] }
  → risk_flags: ["Notion native AI ships in 60 days (estimated)",
                 "Chrome Web Store review hostile to AI extensions Q1"]
  → recommendation: "iterate — narrow wedge to export formats Notion AI doesn't ship"

Returns one of three states:

  • type: xray_submitted + a progress field if the run is still in flight.
  • type: xray_result with the full report once complete.
  • An error envelope if submission was invalid or credits ran out.

Step 5 — Make it part of your workflow

Drop a .cursorrules file in your repo so the agent reaches for Fluenta automatically when you riff on product ideas, instead of making you ask each time:

.cursorrules
# .cursorrules — give the agent permission to use Fluenta proactively

You have access to the Fluenta MCP server. Use it like this:

- For any rough product idea the user is exploring, call
  fluenta_idea_x-ray_sandbox FIRST. It is FREE and instant.
  Surface the verdict (prime / promising / experimental / weak)
  and the top 3 risk flags.

- Only call fluenta_idea_x-ray (the paid run) when the user
  explicitly says "run a full X-Ray" or similar. Confirm the
  2,000-credit cost before submitting.

- After submitting a paid run, return the analysis_id and tell
  the user the run is async (~40 min). Offer to check progress
  later when they ask.

- When the user asks "how's the run going?", call the same tool
  with { analysis_id } to fetch progress; when complete, return
  the structured xray_result object inline.

Tweak to taste. The defaults above bias toward “always run the free sandbox first; only spend credits on explicit go-aheads.” If you want the agent to spend more freely, drop the confirmation step.

Troubleshooting

  • “Tool not found” in Cursor. Restart Cursor. MCP server registration only refreshes on app start.
  • 401 Unauthorized. Key revoked or has a typo. Regenerate at /app/settings/api-keys.
  • 403 Forbidden on the paid tool. The key has read scope only. Issue a read_write key.
  • 402 Payment Required. Out of credits. Free tier ships with 2,000 (one full run); top up to keep going.
  • Run stuck at progress=0 for >5 min. Backend is at peak load — expected occasionally. The run is queued; come back in 30 min.

What to do next

  • Add Fluenta to Claude Desktop too — same config format, paste once and your agent (any agent on the same machine) gets idea-validation tools.
  • Pair with the REST tutorial: run a daily Python job that auto-bookmarks high-LRS ideas, then rerun X-Rays from inside Cursor on whichever ones you want to dig deeper on.
  • Read X-Ray async lifecycle for the wire-level shape if you want to call the same flow from a non-MCP agent.