MCP Task Server

Project for Programming IV Extraclass Assignment 1. It implements a Model Context Protocol (MCP) server that manages tasks in a local JSON file and exposes one resource, two tools, and a prompt.

Capabilities

• Resource tasks://pending: returns only incomplete tasks.

• Tool add_task: adds a task with name, description, and priority (alta, media, or baja).

• Tool complete_task: marks an existing task as completed by ID.

• Prompt daily_summary: builds a template with the current status to generate a daily summary.

The server uses stdio transport. The host or client starts it as a subprocess and MCP messages travel through stdin and stdout.

Related MCP server: TaskTrek MCP Server

Requirements

• Python 3.11 or higher.

• Access to a terminal.

• Claude Desktop, MCP Inspector, or another compatible MCP client for integration.

Installation

Windows (PowerShell)

cd C:\RUTA\mcp-todo-server
py -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -r requirements.txt

macOS or Linux

cd /RUTA/mcp-todo-server
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt

Execution

The stdio server waits for a client to send the first message; therefore, when run directly it does not print a menu or open a port:

python server.py

Use Ctrl+C to stop it.

Demo client

client_demo.py is a compatible MCP client built with the official SDK. It connects via stdio, discovers the capabilities, and runs all the operations requested in the exercise:

python client_demo.py --reset

To save a UTF-8 transcript:

python client_demo.py --reset --output evidence/client_demo_output.txt

The --reset option restores the two initial records before the test. The demo queries the Resource, adds two tasks, completes one, gets the Prompt, and queries the Resource again.

Automated tests

python -m pytest -q

The suite checks:

1. Discovery of the Resource, Tools, and Prompt.

2. Exclusive reading of pending tasks.

3. Creation and completion of tasks.

4. Dynamic content of daily_summary.

5. Error handling when completing a nonexistent ID.

MCP Inspector

With Node.js installed, the official interface can be opened:

npx @modelcontextprotocol/inspector .venv/Scripts/python.exe server.py

On macOS or Linux, change the executable to .venv/bin/python.

Claude Desktop configuration

1. Open claude_desktop_config.example.json.

2. Replace C:\\RUTA\\mcp-todo-server with the absolute path of the project.

3. Copy the mcpServers entry to the Claude Desktop configuration file. On Windows it is usually located at: %APPDATA%\Claude\claude_desktop_config.json.

4. Completely close Claude Desktop and reopen it.

Example:

{
  "mcpServers": {
    "todo-server": {
      "command": "C:\\RUTA\\mcp-todo-server\\.venv\\Scripts\\python.exe",
      "args": ["C:\\RUTA\\mcp-todo-server\\server.py"]
    }
  }
}

Do not use relative paths: the host may start the server from another directory.

Structure

mcp-todo-server/
  server.py
  client_demo.py
  tasks.json
  requirements.txt
  claude_desktop_config.example.json
  tests/
    test_server.py
  diagramas/
    arquitectura_mcp_ecommerce.mmd
    flujo_consulta_inventario.mmd
  evidence/
    client_demo_output.txt
    01_inspector_connected.png
    02_add_task_result.png
    03_complete_task_result.png
    04_resource_pending.png
    05_daily_summary_prompt.png

Publishing to Git

git init
git add .
git commit -m "Implementar servidor MCP de tareas"
git branch -M main
git remote add origin URL_DEL_REPOSITORIO
git push -u origin main

The final repository URL must be added to the report before submitting.

Technical references

• MCP Python SDK

• MCP Specification

• MCP Inspector