MCP-Powered Investigation Agent
Built an AI agent using Model Context Protocol that analyzes video call recordings to surface behavioral patterns and clinical insights.
Technologies Used
Key Features
Project Overview
Designed and built an MCP-powered AI agent that processes video call recordings to identify behavioral patterns, providing structured insights for clinical review.
The Challenge
A mental health platform had thousands of recorded sessions but no scalable way to extract structured insights. Manual review was time-intensive and inconsistent. They needed automated analysis that clinicians could trust.
What I Did
Built a Model Context Protocol server that gives Claude structured access to call transcripts and metadata:
- MCP Architecture: Implemented resources for call data access and tools for analysis operations, keeping the agent’s capabilities well-defined
- Multi-Pass Pipeline: Designed a three-stage analysis: transcription cleanup, behavioral pattern detection, and structured summary generation
- Privacy Design: All processing happens within the client’s infrastructure. No call data leaves their environment
- Batch Processing: Built a queue system for processing historical recordings alongside real-time analysis of new calls
Outcome
The system processes calls in under 2 minutes each, producing structured reports that clinicians can review in a fraction of the time manual analysis required. The MCP architecture means the agent’s capabilities can be extended without rebuilding the core system.