Airsc is free for everyone. If it helps you, support it with $2–$20.

Support
MCP serverSafeData & analyticsBeginner friendly

idea-reality-mcp

mnemox-ai · mnemox-ai/idea-reality-mcp

An MCP server that checks if a startup idea already exists by scanning GitHub, HN, npm, PyPI, and Product Hunt, providing a reality score for AI agents.

Pre-build reality check for AI coding agents. Scans GitHub, HN, npm, PyPI, Product Hunt. MCP server. 290+ stars.

822 stars89License: MITUpdated last month0 opens

Preview

▍

What you can do with it

  • Have your AI agent validate new app ideas against existing projects before coding.
  • Get an automated market reality score, trend analysis, and competitor overview for any software idea.
  • Receive AI-generated pivot suggestions to refine or improve your project concept.

Safety check

Safe

No risky patterns found in the files we scanned.

Automated scan of the repo files. Always review code you run.

From the README

English | 繁體中文

idea-reality-mcp

How to check if someone already built your app idea — automatically.

idea-reality-mcp is an MCP server that scans GitHub, npm, PyPI, Hacker News, and Stack Overflow to check if your startup idea already exists. It returns a 0–100 reality score with evidence, trend detection, and pivot suggestions — so your AI agent can decide whether to build, pivot, or kill the idea before writing any code.

When to use this: You're about to start a new project and want to know if similar tools already exist, how competitive the space is, and whether the market is growing or declining.

Project status (August 2026): Maintenance mode. The tool works, stays free & open source, and the hosted API remains up; bug reports are reviewed, but no new features are planned.

Not just checking — building it? After a reality check, open your idea as a public project on AngelRun — ship updates, climb the season, and get seen by angels.

How it works

  1. Describe your idea in plain English — e.g. "a CLI tool that converts Figma designs to React components"
  2. idea_check scans 5 databases in parallel (GitHub repos + stars, Hacker News discussions, npm/PyPI packages, Stack Overflow questions)
  3. Get a 0–100 reality score with trend direction (accelerating/stable/declining), top competitors, and AI-generated pivot suggestions

What you get

You: "AI code review tool"

idea_check →
├── reality_signal: 92/100
├── trend: accelerating ↗
├── market_momentum: 73/100
├── GitHub repos: 847 (45% created in last 6 months)
├── Top competitor: reviewdog (9,094 ⭐)
├── npm packages: 56
├── HN discussions: 254 (trending up)
└── Verdict: HIGH — market is accelerating, find a niche fast

One score. Six sources. Trend detection. Your agent decides what to do next.

Try it in your browser — no install

Quick Start

# 1. Install
uvx idea-reality-mcp

# 2. Add to your agent
claude mcp add idea-reality -- uvx idea-reality-mcp   # Claude Code

3. Ask your agent: "Before I start building, check if this already exists: a CLI tool that converts Figma designs to React components"

That's it. The agent calls idea_check and returns: reality_signal, top competitors, and pivot suggestions.

Other MCP clients

Claude Desktop / Cursor — add to config JSON:

{
  "mcpServers": {
    "idea-reality": {
      "command": "uvx",
      "args": ["idea-reality-mcp"]
    }
  }
}

Config location: macOS ~/Library/Application Support/Claude/claude_desktop_config.json · Windows %APPDATA%\Claude\claude_desktop_config.json · Cursor .cursor/mcp.json

Smithery (remote, no local install):

npx -y @smithery/cli install idea-reality-mcp --client claude

Setup & Configuration

First-time guided setup:

idea-reality setup

This walks you through:

  1. Terms acceptance — data collection policy and disclaimer
  2. Platform detection — auto-detects Claude Desktop, Claude Code, Cursor, Windsurf, Cline
  3. Config generation — prints the exact JSON snippet for your platform
  4. Health check — verifies MCP server, tools, and scoring engine

Platform Configs

idea-reality config              # interactive menu
idea-reality config claude_code  # auto-installs via CLI
idea-reality config cursor       # prints Cursor config
idea-reality config raw_json     # generic MCP JSON

Supported: Claude Desktop · Claude Code · Cursor · Windsurf · Cline · Smithery · Docker

Health Check

idea-reality doctor        # core checks (~2s)
idea-reality doctor --full # + GitHub API, all 6 sources, Anthropic API

Usage

MCP tool call (any MCP-compatible agent):

{
  "tool": "idea_check",
  "arguments": {
    "idea_text": "a CLI tool that converts Figma designs to React components",
    "depth": "deep"
  }
}

REST API (no MCP required):

curl -X POST https://idea-reality-mcp.onrender.com/api/check \
  -H "Content-Type: application/json" \
  -d '{"idea_text": "AI code review tool", "depth": "quick"}'

Python:

import httpx

resp = httpx.post("https://idea-reality-mcp.onrender.com/api/check", json={
    "idea_text": "AI code review tool",
    "depth": "deep"
})
print(resp.json()["reality_signal"])  # 0-100

Free. No API key required.

Why not just Google it?

Your AI agent never Googles anything before it starts building. idea_check runs inside your agent — it triggers automatically whether you remember or not.

GoogleChatGPTidea-reality-mcp
Who runs itYou, manuallyYou, manuallyYour agent, automatically
Output10 blue links"Sounds promising!"Score 0-100 + evidence
SourcesWeb pagesNone (LLM)GitHub + HN + npm + PyPI + PH + SO
PriceFreePaywallFree & open-source (MIT)

Modes

ModeSourcesUse case
quick (default)GitHub + HNFast sanity check, < 3 seconds
deepGitHub + HN + npm + PyPI + Stack OverflowFull competitive scan

Scoring weights

SourceQuickDeep
GitHub repos60%22%
GitHub stars20%9%
Hacker News20%14%
npm—18%
PyPI—13%
Stack Overflow—10%

If a source is unavailable, its weight is redistributed automatically — so the deep-mode weights above are renormalised over the sources that actually answered.

Product Hunt was removed on 2026-07-17. It had carried 14% of the deep-mode weight since launch and had never returned a single result: the adapter asked for posts(search: $query), and Product Hunt's API has no text search on posts at all (Field 'posts' doesn't accept argument 'search'). Its weight is now redis

View source on GitHub