AgentGuard discovers the AI agents living in your codebase — LangChain, custom, CI, MCP-connected — maps their credentials, tools, and permissions, and hands you the security findings. Before an incident does it for you.
Runs 100% locally. Your code, configs, and credentials never leave your machine.
Agents don't announce themselves. They're a Python script with an API key, a service account, an MCP config on a laptop, a nightly CI job. AgentGuard finds them and shows what they can reach.
Detects agent frameworks, model SDKs, MCP servers and configs, CI agents, and AI CLIs across your repositories — including the ones nobody registered.
Builds the security graph: agent → credential → MCP server → tool → system. Shared keys, destructive tools, and toxic permission combinations light up.
Ranked findings with evidence and a concrete fix — rotate this key, gate that refund tool, assign this orphaned agent an owner.
Two commands. It reads your files, writes an HTML report next to you, and makes no network calls — you can run it with the wifi off.
Prefer not to install anything? Read the sample report — it's the real output from a test environment. Found something surprising in your own scan? Tell us — that's exactly what we want to hear about.
Asking a security team to hand a young vendor read access to everything is a six-month conversation. So we didn't build that. The scanner runs in your environment, the report is a file you own, and nothing — no code, no credentials, no telemetry — leaves your machine. When you want continuous monitoring, that's a conversation we earn after the first report.
Posture first. Trust before enforcement.
Inventory every agent, credential, MCP server, and tool.
Audit trail of what each agent actually did.
Policies for what each agent may do.
One policy, compiled to every gateway you already run.
Anomaly detection and automatic suspension.
We're onboarding a small group of design partners: you run the scanner, we walk the findings with you, your feedback shapes the product. No data leaves your environment at any point.