The Problem

Large Language Models (LLMs) like Claude are powerful reasoning engines, but they operate with a fundamental limitation — they are isolated from live, real-world data. When a user asks Claude about today’s weather, stock prices, or any other live information, Claude can only respond from its training data, which has a knowledge cutoff date.

Specifically, the following pain points exist in standard LLM deployments:

  • Claude has no access to real-time weather data from external APIs
  • Users must manually look up weather and paste it into the chat — poor UX
  • No standardized way for Claude Desktop to call external tools or APIs
  • Developers had no simple framework to extend Claude with custom capabilities
  • Enterprises cannot integrate proprietary data sources into Claude conversations

This creates a significant gap between what Claude can know (static, trained knowledge) and what users actually need (live, contextual, real-world data).

 Our Solution

We built a Weather MCP Server — a lightweight Node.js service that implements Anthropic’s Model Context Protocol (MCP). This server acts as a bridge between Claude Desktop and the OpenWeatherMap API, giving Claude the ability to fetch live weather data on demand.

The solution works as follows:

  • A TypeScript-based MCP server exposes a get_weather tool to Claude Desktop
  • Claude Desktop auto-starts the server via stdio transport (no manual startup needed)
  • When a user asks about weather, Claude calls the tool with the city name
  • The server fetches live data from OpenWeatherMap API and returns structured results
  • Claude presents the weather data naturally in the conversation

Key Differentiators:

  • Zero manual steps — server starts automatically with Claude Desktop
  • Type-safe implementation using TypeScript and Zod schema validation
  • Clean separation of concerns — MCP handles communication, API handles data
  • Easily extensible — new weather tools (forecast, alerts) can be added in minutes
  • Production-ready logging and error handling built in

 Solution Architecture

The architecture follows a clean 3-layer design:

Layer 1Claude DesktopAI Client / UILayer 2Weather MCPNode.js / TypeScriptLayer 3OpenWeatherExternal REST API

Communication Flow:

  • User types weather query → Claude Desktop receives it
  • Claude identifies get_weather tool is needed → sends tools/call over stdio
  • MCP server receives JSON-RPC message → extracts city parameter
  • Server calls OpenWeatherMap REST API with city + API key
  • API returns JSON → server formats and returns structured text
  • Claude Desktop receives result → presents weather to user naturally

Configuration

The server is registered in claude_desktop_config.json and auto-launches on Claude Desktop startup. No manual intervention required after setup.

RELATED ARTICLESMORE FROM AUTHOR