Claude Desktop just got the ability to plan wine trips. Not hallucinated wine trips — real itineraries based on GPS-mapped wineries across 18 regions, structured around actual harvest seasons and tasting appointment culture. That's what shipping our MCP server means in practice.
## What Is MCP, Exactly?
Model Context Protocol (MCP) is an open standard that lets AI assistants like Claude connect to external data sources at query time. Instead of the model guessing what's in Napa Valley — drawing on whatever training data it absorbed — it can query our actual database. The difference is the same as the difference between a travel agent who visited the region last month and one who last visited in 2023.
MCP servers expose tools. Claude decides which tool to call based on the user's question. The server runs the query, returns structured data, and Claude uses that data to compose its response. No hallucinations. No fabricated winery names. No plausible-sounding addresses that don't exist.
## What Our MCP Server Does
Our server exposes four tools:
- **get_region_info** — Returns the full region guide for any of our 18 supported wine regions: geography, climate, dominant varietals, harvest calendar, tasting culture, and logistics notes. - **search_wineries** — GPS-based winery search filtered by region and name. Returns real coordinates, visiting hours, and booking requirements. - **plan_trip** — Takes a region, duration, and travel dates and returns a structured day-by-day itinerary grounded in actual geography and seasonal context. - **list_regions** — Enumerates all 18 available regions so Claude knows what it can and can't help with.
The server uses stdio transport and zero external dependencies — Node 18+ built-ins only. It runs locally on the user's machine as a sidecar process that Claude Desktop manages automatically.
## Why This Changes the Output Quality
Here's the concrete difference. When a user asks Claude "plan me a 4-day Alsace wine trip in October," Claude can now:
1. Query our region data to pull accurate sub-AVA information, harvest timing for Alsace in October, and which estates require advance booking during harvest season 2. Generate a day-by-day itinerary grounded in real geography — routes that make sense, estates that are actually near each other 3. Return accommodation suggestions enriched with affiliate context, because we know which villages are logistically central for each itinerary
Without MCP, Claude would produce a confident-sounding itinerary that any experienced wine traveller would immediately recognise as generic. With MCP, the output is grounded in the same data our editorial team has spent years building.
## The Architecture
The MCP server lives at `src/mcp-server/index.ts` in the TheWineTrip codebase. We'll publish it as `thewinetrip-mcp-server` on npm so users can install it with a single command and a one-line config change.
Users add it to their Claude Desktop `config.json`:
```json { "mcpServers": { "thewinetrip": { "command": "npx", "args": ["-y", "thewinetrip-mcp-server"] } } } ```
Restart Claude Desktop, and the tools are available for any wine trip planning conversation.
## What's Next
The stdio transport works for local Claude Desktop use. Next we're building a remote HTTP MCP server using SSE transport, which will make the same tools available to web-based Claude use cases — no local install required. We're also expanding region coverage and deepening the integration with VineStays for real accommodation booking rather than suggestions.
The goal is a Claude that can take someone from "I want to do a wine trip" to a bookable itinerary in a single conversation, with every recommendation grounded in data we stand behind.
---
**Install the MCP server today** and let Claude plan your next wine trip with real data behind it.