So I built an MCP server that does the searching before you write code. It scans 5 real-time sources (GitHub, Hacker News, npm, PyPI, Product Hunt) and returns a quantified reality_signal (0-100) with actual evidence — repo counts, star counts, top competitors, and pivot suggestions.
Example: "AI code review tool" → reality_signal: 90, 847 repos, top competitor reviewdog (9,094 stars), 254 HN mentions.
What it's NOT: not a business plan generator, not an LLM opinion wrapper. Every number comes from a real API call you can verify.
- Quick mode: GitHub + HN (default) - Deep mode: all 5 sources in parallel - Works with Claude Desktop, Cursor, Claude Code - One-line install: uvx idea-reality-mcp - Also available as web demo (no install needed)
GitHub: https://github.com/mnemox-ai/idea-reality-mcp Web demo: https://mnemox.ai/check MCP Registry: io.github.mnemox-ai/idea-reality-mcp
Built with Python + FastMCP, 120 tests, published on PyPI. Happy to answer any questions about the scoring algorithm or MCP integration.