MCP Security Orchestrator

Voice-controlled multi-agent security platform that orchestrates AI agents via AWS Bedrock MCP to scan targets, analyze security flaws, and generate reports

Comment

Inspiration

Security assessments traditionally require specialized expertise and manual tool coordination. I envisioned a platform where anyone could simply speak their security assessment needs and have AI agents autonomously collaborate to deliver comprehensive results. The release of AWS Bedrock's Model Context Protocol (MCP) provided the perfect foundation to build truly interoperable AI agents that can share context and coordinate complex security workflows.

What it does

MCP Security Orchestrator is a voice-controlled multi-agent security platform featuring:

  • Voice Interface: Speak commands naturally - "Scan owasp-juice.shop and generate a security report"
  • Multi-Agent Orchestration: AI agents collaborate autonomously - Security Auditor scans targets while Report Generator creates professional PDFs
  • Real-time Dashboard: Watch agents think, call tools, and produce results in a visual Streamlit interface
  • MCP Integration: Security tools exposed via Model Context Protocol for seamless agent-tool interaction
  • Comprehensive Scanning: DNS lookup, port scanning, security headers analysis, cookie auditing, CORS policy checks, and more
  • Retool Integration: Automated report delivery via email through Retool agent workflows
  • TinyFish MCP: Additional tool capabilities through TinyFish MCP server integration

How I built it

Architecture:

  • AWS Bedrock with Claude 3.5 Sonnet as the AI backbone
  • Model Context Protocol (MCP) server exposing security tools (nmap, DNS, HTTP analysis)
  • TinyFish MCP for extended tool capabilities and agent-to-agent communication
  • Retool for automated report generation workflows and email delivery
  • Streamlit for the interactive dashboard UI
  • ElevenLabs for text-to-speech and speech-to-text voice features
  • Multi-Agent Orchestrator that plans and delegates tasks across specialized agents

Key Components:

  1. SecurityAuditorAgent - Performs security scans using MCP tools
  2. ReportGenerationAgent - Creates detailed PDF reports from scan data
  3. MultiAgentOrchestrator - Coordinates agent collaboration and task delegation
  4. VoiceService - Handles speech recognition and synthesis
  5. RetoolAgent - Handles report formatting and email delivery workflows

Challenges I ran into

  1. Python Closure Scoping: Nested async functions in Streamlit caused UnboundLocalError with the json module. Solved by careful scope management with local imports and variable aliasing.

  2. Real-time Streaming: Displaying agent thinking, tool calls, and results in real-time required careful placeholder management in Streamlit's execution model.

  3. MCP Authentication: Integrating AWS SigV4 authentication with MCP's streamable HTTP transport required custom transport implementation.

  4. Voice Integration: Synchronizing voice input/output with async agent execution while maintaining responsive UI.

  5. Multi-MCP Coordination: Connecting multiple MCP servers (security tools, TinyFish, Retool) required careful session management and authentication handling.

Accomplishments that I'm proud of

  • Seamless Voice Control: Users can speak naturally to trigger complex multi-step security assessments
  • Real-time Agent Visualization: Watch AI agents think, reason, and collaborate in real-time through the dashboard
  • End-to-end Automation: From voice command to PDF report email delivery without manual intervention
  • MCP-native Architecture: Built from the ground up using AWS Bedrock's Model Context Protocol for true agent interoperability
  • Multi-MCP Integration: Successfully orchestrated multiple MCP servers (custom security tools, TinyFish, Retool) in a unified platform
  • Production-ready UI: Professional Streamlit dashboard with dark theme, real-time updates, and intuitive controls

What I learned

  • Deep understanding of AWS Bedrock's MCP implementation and agent orchestration patterns
  • Streamlit's async execution model and session state management
  • Building production-ready voice interfaces with ElevenLabs
  • Designing multi-agent systems where agents can delegate to and collaborate with each other
  • Handling complex authentication flows (Auth0, AWS SigV4) in distributed agent architectures
  • Integrating third-party MCP servers (TinyFish, Retool) with custom agent workflows

What's next for MCP Security Orchestrator

  • Add more security tools (Nikto, SQLMap integration)
  • Implement continuous monitoring mode with scheduled scans
  • Add team collaboration features and shared dashboards
  • Expand voice commands with more natural language understanding
  • Integration with ticketing systems (Jira, ServiceNow) for vulnerability tracking
  • Deeper Retool integration for customizable reporting templates

Built With

  • asyncio
  • auth0
  • aws-bedrock
  • aws-sigv4
  • claude-3.5-sonnet
  • elevenlabs
  • model-context-protocol-(mcp)
  • nmap
  • python
  • retool
  • streamlit
  • tinyfish-mcp
Share this project:

Updates