Xonta

Xonta is a self-hosted trust layer that sits in front of any MCP server, giving you verifiable proof of what your agents actually did. Key capabilities: • Signed receipts for every tool call and spend • Budget limits and per-call risk caps • Removal of hidden instructions from tool output • Receipt verification with a public key and plain arithmetic • Open source under Apache-2.0, one command to run An agent's own report is not reliable evidence, especially under pressure. Xonta's gateway chains each signed receipt to the one before it, so any edit, deletion, or reordering becomes detectable. Budgets and exposure caps stop risky calls before they execute, and output filtering removes hidden instructions that could hijack an agent. Because verification needs only a public key and simple math, auditors, customers, and other orchestrators can check results without trusting Xonta itself. It is ideal for teams running agents in production, security and compliance staff, and developers who need spending control and an auditable trail of tool activity.