Hey Product Hunt, AI coding agents are moving from suggestions to real actions: reading files, querying databases, running shell commands, calling internal APIs, and touching deployment workflows through MCP. That is powerful, but it creates a security gap. Most teams can identify the agent, but they still need a control point for the action itself: which tool, which arguments, which environment, which credential, and whether a human should approve it before it runs. AgentGate is our answer to that gap. It sits between MCP-compatible agents and the servers/tools they call. Every tool call is checked before execution, then allowed, denied, or routed to a request-bound human approval. Each decision is written to an audit trail so security and platform teams can inspect what happened later. The first version focuses on AI coding agent workflows: - default-deny policy for risky MCP calls - human approval for sensitive actions - gateway-held upstream credentials - secret redaction and SSRF guardrails - audit evidence for security reviews - hosted, local connector, and self-hosted deployment paths We are launching this early because teams are adopting agents faster than their authorization model is evolving. I would love feedback from security, platform, DevOps, and AI tooling teams: Where do you think the approval boundary should sit for AI agents that can touch production systems?

AgentGate Labs

Approve AI agent tool calls before they run

Approve AI agent tool calls before they run