Model Context Protocol (MCP) Server & Agent

A simple implementation of a Model Context Protocol server that manages context data for AI models.

Setup

1. Install the required dependencies:

uv venv
uv pip install -r requirements.txt

2. Run the server:

uv run server.py

The server will start on localhost:8080.

API Endpoints

Store Context

curl -X POST http://localhost:8080/mcp/context \
  -H "Content-Type: application/json" \
  -d '{
    "context_id": "example-context",
    "data": {
      "key": "value",
      "metadata": "example"
    }
  }'

Get Context

curl -X GET http://localhost:8080/mcp/context/example-context

Delete Context

curl -X DELETE http://localhost:8080/mcp/context/example-context

list all the tools

curl -X GET http://localhost:8080/mcp/tools

Run the agent

OPENAI_API_KEY=$OPENAI_API_KEY uv run agent.py 

Sample output:

Connected to server
LLM decided to use: calculator
Reasoning: The user is asking for the sum of 5 and 3, which is a basic arithmetic operation suitable for the calculator tool.
Tool 'calculator' executed with result: {'status': 'success', 'result': 8}

Final result: {'status': 'success', 'result': 8}
LLM decided to use: calculator
Reasoning: The user requested to multiply two numbers, so a calculator tool is appropriate to perform this arithmetic operation.
Tool 'calculator' executed with result: {'status': 'success', 'result': 24}

Final result: {'status': 'success', 'result': 24}
Disconnected from server