Doubt as a superpower: the MCP that questions every move AI makes

Descartes-MCP is a weird hybrid between a philosophical sparring partner and a paranoid project manager. The idea is simple but clever: when Claude drafts a plan, this tool doesn't just wave it through like you normally would. Instead, it drags Claude back over every single step and asks: "Really? What's your evidence?"

The mechanism is recursive—kind of obvious from the name. It's not a one-time check. If Claude's answer lacks solid ground—say, "I'll do X because it's logical" with zero data backing it—Descartes stops him cold and insists: bring receipts, or admit you don't have any. It keeps going until it hits convergence, meaning every decision is anchored to something actually verifiable.

Here's the plot twist: you don't need expensive API keys, extra cloud services, or complicated infrastructure. It just works with Claude. It's like having a relentlessly pedantic colleague living inside your model who simply won't let you move forward without proof. And honestly? In a world where AI hallucinations are still a genuine problem, a little pedantry might actually save your skin.

The developer describes it as the "sibling" of another project called "The Asker," which runs on the same philosophy: methodical questions until clarity clicks. It's all open source too, which means you can peek under the hood and see how it actually works—and tweak it if your own paranoia has special requirements.

Practical value? If you're building something where wrong calls have real costs (spoiler: they always do), this kind of tool cuts the risk of moving forward on sand dramatically. Not a magic bullet, but it's like having a relentless quality controller who never clocks out.