ClimaSafe · MCP server

ClimaSafe as a
Model Context Protocol server.

The ClimaSafe risk engine — prediction, personal profiles, routines and risk charts — exposed as an MCP server so any AI assistant can ask about heat and cold risk directly from its own interface.

Telegram bot github.com/ANFAIA/ClimaSafe ← back to project
01
what it is An MCP for heat and cold risk
The ClimaSafe MCP server exposes the same prediction engine as the web and the Telegram bot, through the Model Context Protocol. An assistant configured with it can fetch personal cardiovascular risk predictions, manage user profiles and routines, and render the hourly risk chart, all without leaving the chat.
Tools
risk prediction · risk chart · profiles · routines
Access
read-only by default · writes need a token
Transport
stdio (local) · SSE (remote)
02
connect Add it to Claude Desktop, opencode or Cline
Add the server to your opencode.json, .mcp.json or equivalent MCP configuration. It runs with uv, so the project environment is resolved automatically: 
{
  "mcpServers": {
    "climasafe": {
      "command": "uv",
      "args": ["run", "python", "-m", "agents.tools.prediction_mcp_tool", "--stdio"]
    }
  }
}
Run it from the root of the ClimaSafe repository. On Claude Desktop, paste the same block into claude_desktop_config.json under mcpServers.
03
permissions Read-only by default, writes behind a token
Sin token, solo lectura. The server starts in read-only mode when no write token is set. Write operations — creating or deleting profiles, managing routines, setting daily alerts — require CLIMASAFE_MCP_WRITE_TOKEN to be present in the environment. Predictions and risk charts work without it.
Source, tools and the full component map live in the repository: github.com/ANFAIA/ClimaSafe — see documentacion/componentes.md for how the MCP server fits into the system.