Secure-Ai-Agents-Mcp-Courses

Produce and evaluate a review-ready security package for an AI agent with MCP tools.

Run a controlled indirect prompt injection, observe an unsafe policy, add enforcement, and verify the blocked action.

Map agent authority, data origins, side effects, and approval points before connecting tools.

Design MCP tools, authorization, input checks, and transport boundaries for least-privilege agent access.

Log agent decisions, detect policy failures, stop unsafe runs, and recover with evidence.

Test your understanding of agent authority, MCP tools, prompt injection, approval, logging, and recovery.