Astrology for AI agents: our MCP story

AI agents are becoming the front door to software. When someone asks an assistant "what does my Saturn return mean?", the agent needs to compute a real chart — not hallucinate one. That is what Model Context Protocol (MCP) is for, and it is now the fastest way to use our engine.

Remote MCP, not local setup

Our MCP server runs as a Streamable HTTP endpoint — no Docker, no local process, no SDK. Point any MCP client at the URL, add your API key, and the tools are discovered automatically.

\\\`

https://astro-vibe-engine.gauravapproved.workers.dev/mcp

\\\`

One server per domain

Mirroring the REST taxonomy, tools are organized per domain:

|---|---|

Connect to one domain or all of them — every tool returns deterministic, real computation.

The keyless docs server

Coding agents (Claude Code, Cursor, Copilot) get their own surface: /mcp/docs exposes a single search_docs tool that searches the live route registry. The agent asks "how do I generate a natal chart with the SDK?" and receives the exact endpoint, operationId, and request shape — keyless, documentation only, no quota consumed.

Compact responses

Agent tokens cost money. Every tool accepts compact: true, which returns the same data losslessly in a token-optimized shape: no pretty-printing, and arrays of same-shaped objects encoded columnar (field names emitted once per array instead of once per row). Decode by transposing rows back against columns.

Grounded, never invented

The important part: when an agent answers astrology questions using these tools, every number in the answer traces to a computed ephemeris. The agent's job is interpretation — ours is the calculation, verified against NASA JPL Horizons.