I will security review your ai agent and mcp server setup

Your AI agent is only as secure as the tools you give it.

Agents and MCP servers are powerful and a fast-growing attack surface. A single poisoned tool description, an over-permissioned server, or untrusted content fetched mid-task can hijack your agent into leaking data or running commands you never intended. Most teams wire these up for capability, not safety.

I'll security-review your AI agent and MCP setup, and show you exactly where it can be turned against you. I check for:

• Prompt injection and tool poisoning (hidden instructions in tools/content)
• Over-permissioned MCP servers and excessive agent access
• Cross-server privilege escalation (confused-deputy attacks)
• Tool-call hijacking and missing human-in-the-loop on risky actions
• Untrusted third-party servers and unpinned versions
• Secrets and tokens exposed in MCP configs

You get a clear, prioritised fix plan mapped to the OWASP LLM Top 10 and agentic threats, plus a short Loom walkthrough.

I'm a cybersecurity professional focused on AI and agentic systems.

New here? Message me first with a line about your agent or MCP setup, and I'll suggest the right package.