Built a spending mandate layer for AI agents — set limits once, agent can't overspend

**Adding the Github and install details in the comment, any input is greatly appreciated!** Been building autonomous agents and kept running into the same problem: once you give an agent access to spend money, there's nothing stopping it from going way over budget or hitting merchants it shouldn't. So I built MCP server that acts as an authorization gate before every transaction. You define the mandate once: \- Max per transaction \- Daily/weekly spending cap \- Allowed merchants only Then every time the agent wants to spend, it calls authorize\_purchase first. Approved = go ahead. Denied = agent stops and reports back. Useful for: \- Personal assistant agents with a nightly budget \- Autonomous research agents capped per run \- Any workflow where you want spending guardrails without approving every action Every decision is logged so you can audit exactly what the agent tried to do.