LLMs are encouraging by nature. Ask Claude or GPT "should I build X?" and it will say yes. io.github.mnemox-ai/idea-reality-mcp injects actual data into that conversation.
How It Works
Describe a software idea in natural language. The server searches five real databases in parallel — GitHub repositories, Hacker News discussions, npm packages, PyPI packages, and Product Hunt launches — and returns a reality signal score from 0–100 indicating market saturation, ranked competitors with URLs and star counts, and actionable pivot suggestions.
Two search modes: Quick (GitHub + Hacker News, fast) and Deep (all five sources, comprehensive). The scoring uses transparent, published weights — not ML, not a black box. Quick mode: GitHub repos 60%, stars 20%, HN 20%. Deep mode: repos 25%, stars 10%, HN 15%, npm 20%, PyPI 15%, Product Hunt 15%. You can see exactly why your idea scored what it scored.
The Details
Claude Haiku extracts optimal search keywords from natural language descriptions, with a dictionary-based fallback (150+ term mappings, including Chinese) if the API is unavailable. A companion GitHub Action (mnemox-ai/idea-check-action@v1) auto-validates any issue labeled "proposal" — making this literally a linter for project ideas in your CI pipeline.
Zero-cost local execution. No paid API required — GitHub and Product Hunt tokens are optional for higher rate limits.
Who Built It
Mnemox AI is a small Taipei-based team that openly lists an AI (Claude) as their CIO. Their thesis: "AI shouldn't start from zero every conversation." Their other project, tradememory-protocol (47 stars), applies the same philosophy to AI trading memory. 219 stars and 17 forks in five days. Eight releases (v0.1.0 through v0.4.0) in the same window — they're iterating fast.
Why It's Interesting
This represents a maturing of the MCP ecosystem. It is not "give the AI access to X." It is "make the AI smarter about Y." The server changes agent behavior at the decision-making level — intercepting the moment between "I have an idea" and "let me start building" with evidence from five real data sources. The CI integration makes this particularly practical: gate project proposals at the issue level, before anyone writes code.
Score: 60. No flags. MIT license.
Sources: Mnemox AI — GitHub · idea-reality-mcp — repo · PyPI · Live demo · Scorecard: io.github.mnemox-ai (score 60)