Miklat MCP: An AI-Powered Bomb Shelter Finder For Israel

I built an MCP (Model Context Protocol) server that helps AI agents find public shelters (miklatim tziburim) in Israel. It runs on Cloudflare Workers and is freely accessible to any MCP-compatible client.

MCP endpoint

https://mcp.jlmshelters.com/mcp

Connect using streamable HTTP transport from Claude Desktop, Claude Code, Cursor, ChatGPT, or any MCP client.

Available tools

The server exposes 9 tools:

• search_shelters — Free-text search across name, address, neighbourhood

• find_nearest_shelters — Find nearest shelters by latitude/longitude

• list_neighborhoods — All neighbourhoods with shelter counts

• get_shelter_by_id — Get a single shelter's full details

• get_stats — Summary stats: total count, type breakdown, capacity

• list_cities — All supported cities

• get_directions_link — Google Maps / Waze navigation links

• filter_shelters — Filter by type, capacity, accessibility

• list_shelters_in_neighborhood — All shelters in a specific neighbourhood

Coverage

Currently covers Jerusalem with 198 public shelters. Data is community-maintained in the Miklat-MCP-Data repository using GeoJSON with a three-stage validation pipeline (raw → structured → validated → production). To add a new city, submit a PR with a GeoJSON file following the documented schema.

How to connect

Add to your MCP client config:

{"mcpServers":{"miklat":{"type":"streamableHttp","url":"https://mcp.jlmshelters.com/mcp"}}}

View the project on GitHub