MCP Server

MCP Proof of Concept

This repository contains a simple Model Context Protocol (MCP) server implemented with FastAPI. The goal is to expose herd data through a discoverable, versioned API that can be deployed to AWS Fargate.

Running locally

1. Install dependencies:The database path can be configured via the DATABASE_PATH environment variable. If not set it defaults to mcp.db inside the working directory.
   pip install -r requirements.txt
2. Seed the SQLite database:
   python -m app.seed
3. Start the API server:
   uvicorn app.main:app --reload
4. Authenticate with the token fake-super-secret-token when calling the API.

The MCP discovery file is available at model_context.yaml.

Using the agent

An agent package is provided to interact with the MCP server. After the server is running you can list the herd data like so:

The agent reads model_context.yaml to discover the API path and returns the JSON response from the server. For full YAML support install the optional PyYAML dependency; otherwise a limited built-in parser is used.

Running tests

Container

A Dockerfile is provided to run the server in a container. Build with:

Terraform

The terraform directory contains a minimal configuration showing how the container could be deployed to AWS (e.g. Fargate). It creates an ECR repository for the image.