Audit your agent's tools before they ship
Paste your OpenAI function specs, Anthropic tools, or an MCP tool manifest. Agent Guard flags excessive agency, code-execution and SSRF sinks, secret access, and data-exfiltration chains — each mapped to the OWASP 2025 LLM Top-10 with a concrete fix. Runs client-side; an API key unlocks the advanced cross-tool engine.
Excessive agency (LLM06)
Destructive, financial, and irreversible actions exposed to the model with no human-approval gate.
Exfiltration chains (LLM02)
A tool that reads secrets plus a tool that can reach the network — an injection-driven exfil path across tools.
Injection sinks (LLM01/05)
Arbitrary code/command execution and unconstrained free-form url/path/query parameters that turn a prompt into an exploit.
Use it in CI
Run Agent Guard on your tool manifest in CI to catch excessive agency before it ships. The browser tool is free and private; the hosted API (with an API key) unlocks the advanced cross-tool engine and saved history.
# analyze an MCP / tools manifest via the hosted API
curl -X POST https://www.opensecureai.com/api/agent-guard \
-H "content-type: application/json" \
-H "authorization: Bearer YOUR_OPENSECUREAI_API_KEY" \
-d @tools.json
# → { "ok": true, "engine": "hosted-advanced",
# "result": { "level": "Critical", "score": ..., "findings": [ ... ] } }