Inspiration

The Haven Project serves thousands of youths across multiple programs with a very small staff, which means an overwhelming amount of participant data. Existing case management tools like Apricot are expensive, complex, and often underutilized, requiring extensive training just to perform basic tasks. We wanted to bridge the gap between ambitious social workers and powerful data systems with a model that can understand both. Hence, our MCP server.

Tech Stack

  • Frontend: React + TypeScript + Tailwind CSS
  • Backend: MCP (Model Context Protocol) server + Python + Go
  • LLM: Groq
  • Database: PostgreSQL (Supabase), Google Cloud
  • Infrastructure: Debian Linux server

Features

  • Natural language: A side popup chatbox allows staff to manage the entire system using plain English
  • Patient data tracking: View individual patient profiles with current metrics and historical case notes
  • Analytics & reporting: Automatically generated summary statistics and trend analysis across all patients alongside an analytics dashboard
  • Flexible data: Add custom fields, history logs, and grant-specific metrics
  • Intake via conversation: Create new patient records directly from intake summaries
  • Opportunities matching: Match job, education, or housing opportunities to the most fitting clients using the assistant
  • Soft risk assessment and visit notes drafting generated by the MCP client
  • Low cost & transparent: Hundreds of LLM requests cost only cents, with no expensive third-party software licenses
  • Scalable & open source: Each nonprofit can fully own and deploy their own instance independently
  • Built a concurrent web scraper that aggregates housing and grant opportunities from public nonprofit and government sources in Boston that is updated to the Supabase database and deployed on GCP with automated 24-hour refreshes.

What's next for Haven Helper

  • Make our own installation package to automatically set up Haven Bridge instances
  • Add authentication and different levels of control for staff, volunteers, and administrators
  • Easy extraction of data analytics to a pdf or some other format

Built With

Share this project:

Updates