What we shipped: Mantrana now speaks MCP — bring the spec council into Claude Code
Hey Product Hunt 👋
Quick update since the September launch. Mantrana started as a website: paste a rough product idea, and a council of four specialist AI jurors (architect, red-teamer, product strategist, operations lead) cross-examines it and files one buildable specification. Today the useful part of that moved to where coding agents can actually use it: Mantrana now ships an MCP server, so the council can live inside Claude Code, ZCode, OpenCode, or Cursor instead of a browser tab.
How it works, end to end:
1. Create an account at mantrana.co and issue an API key (Account menu → Developer & credits; the raw key is shown once).
2. Connect your agent:
claude mcp add --transport http mantrana https://mantrana.co/mcp --header "Authorization: Bearer mk_YOUR_KEY"
3. In your agent, call mantrana_specify with your rough idea. It deliberates in the background (2–6 minutes by tier). mantrana_wait_for_run long-polls the result, then mantrana_get_specification hands back the full spec — schemas, API contracts, threat matrix, phased roadmap. Your agent codes against it immediately instead of wandering.
Every account gets $10 in credits once (including accounts that signed up during launch week — sign in and it's there). Runs cost $1.50 / $2.50 / $5.00 by tier, so $10 is about six MVP-tier specifications, spendable on the website or through the MCP server. Web users keep the 3-run free ladder as before.
The honest numbers, from our published three-task benchmark (methodology on the site): agents working from a Mantrana spec took 84.6% fewer steps on average, cut token spend by ~84% ($64.60 net savings across the three cases), and first-pass build success went from 30.4% to 99.5%. One real user-published example from last week: https://mantrana.co/s/86v3Ti08b3Y — that's the level of output a single MVP-tier run produces.
Full developer docs: https://mantrana.co/developers
If you try it, I'd genuinely like to hear what your agent built (or broke). Feedback on the tool interface especially — it's new.
Replies