Auths Agent Guard
Pre-Execution Spend Budget Caps & Kernel Capability Sandboxing
Pre-execution budget caps, time-to-live bounds, and OS kernel sandboxing for Model Context Protocol (MCP) servers. Refuses overspending before it happens and leaves signed audit receipts.
Agent Guard Execution Gate
Agent requests tool execution with parameters and cost metadata.
Encrypted Touch ID Approvals via Murmur
When an AI agent requests an elevated spend or high-risk capability, it deposits a sealed challenge envelope at an encrypted zero-knowledge relay. You receive an instant Touch ID prompt on your Mac—no inbound open ports, no master keys exposed.
One-Line MCP Client Integration
Prepend auths wrap to any existing MCP server line in your client config. No toolchain or central server required.
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": [
"-y", "@auths-dev/mcp", "wrap",
"--scope", "paid.call",
"--budget", "$50",
"--ttl", "30m",
"--", "npx", "-y", "@modelcontextprotocol/server-filesystem", "/workspace"
]
}
}
}Capability Scope
Restricts tool calls to authorized functions (e.g. paid.call or fs.read). Prevents arbitrary command execution.
Spend Budget Cap
Enforces dollar spend caps ($50). Refuses tool invocation at protocol boundary before overspending happens.
Time-to-Live Bound
Hard time expiration (30m). Automatically invalidates agent session capabilities upon TTL expiration.
Verifiable Spend Receipt
Emits signed DSSE spend receipts verifiable 100% offline via browser WebAssembly without cloud servers.