Roundtable MCP - Give your agent a council of frontier models to debate

Hi Product Hunt 👋 Here's the honest story behind Roundtable. I live in Claude Code. Most of my day is prompting, planning, reviewing what the agent writes. And I kept hitting this moment where the AI would suggest something, an architecture choice, a migration strategy, a security pattern, and I'd just... freeze. Not because the answer was obviously wrong. Because it sounded exactly right and I had no way to pressure-test it. So I'd do what I think a lot of us do: copy the context, paste it into ChatGPT. Then paste it into Gemini. Then maybe Grok. Compare the answers in my head. Try to figure out where they disagree and why. It worked, sort of. But it was slow, messy, and I lost all the context switching between tabs. I kept thinking: why am I the one doing this manually? The models should be talking to each other. Not in parallel silos, but sequentially, where each one reads what the others said and actually responds to it. Challenges it. Builds on it. So we built that. Roundtable is an MCP server. One tool call from your AI agent, and it convenes a council of rival models: Claude, GPT, Gemini, Grok, DeepSeek. Each gets a role (Security Reviewer, Systems Architect, Devil's Advocate). They debate sequentially, each reading and pushing back on what came before. Then a moderator synthesizes the consensus, the dissent, the blind spots, and a verdict. It works inside Claude Code, Cursor, VS Code, Windsurf, Gemini CLI. One config line. 30 seconds to your first council debate. This isn't just a hunch. Multi-agent debate has been validated at ICML (Best Paper 2024, +28pp accuracy), NeurIPS, and ICLR. When models are forced to disagree with each other, they catch things no single model catches alone. We didn't invent the science. We just made it one tool call away. I use Roundtable to build Roundtable. Every architecture decision, every "should we use X or Y," every security review goes through the council. We ship fewer bugs and make better calls. Try it. I'd genuinely love to know: what's the first question you'd put to a council of rival AI models? Also would love feedback on pricing, would you guys prefer subscription or usage based pricing?