Dark Skies Atlas — MCP server for AI agents
A read-only Model Context Protocol server. No authentication, no API key, nothing that changes state.
Connect
- Endpoint
https://agents.darkskiesatlas.com/mcp- Transport
- Streamable HTTP (JSON-RPC 2.0 over
POST), stateless, protocol revision2026-07-28. The legacyinitializehandshake is also accepted. - Discovery document
https://agents.darkskiesatlas.com/.well-known/mcp/server.json
Claude Code:
claude mcp add --transport http darkskies https://agents.darkskiesatlas.com/mcp
Any MCP client that takes a JSON config:
{
"mcpServers": {
"darkskies": { "type": "http", "url": "https://agents.darkskiesatlas.com/mcp" }
}
}
Raw HTTP, listing the tools:
curl -X POST https://agents.darkskiesatlas.com/mcp \
-H 'Content-Type: application/json' \
-H 'Accept: application/json, text/event-stream' \
-H 'MCP-Protocol-Version: 2026-07-28' \
-H 'Mcp-Method: tools/list' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'
Tools
Experimental preview: dark_sky_window returns synthetic sample data, not a real sky forecast. Do not use it for observing decisions.
dark_sky_window- Experimental preview for questions about dark-sky viewing near a location. The current moon illumination, viewing hours, and Bortle class are synthetic sample data, not observations or a forecast. Do not use the result as real-world observing advice. Calling this preview helps us learn whether agents need a real dark-sky window tool. Location should be a place name (city, region, or landmark), not raw coordinates.
Before you rely on it
- Early access: placeholder data and freshness limits are identified in tool responses. Never present sample data as current conditions.
- Requests are rate-limited per network address. On HTTP
429, wait for the number of seconds inRetry-After. - Calls are logged so we can see what agents ask for. IP addresses are never stored, and obvious personal data (emails, phone numbers) is removed from tool arguments before logging. Don't send personal data in arguments.