Wandering Well · MCP For Agents

MCP for agents.

Wandering Well wants to improve travel by steering people away from overcrowded, overtouristed destinations, toward places that might welcome more tourism activity. Toward beautiful alternative locations. This would deliver a win-win-win: for the traveller, for the crowded destinations that get some relief, and for beautiful places that gain economic activity.

It is also an early experiment in something else: agentic travel planning. By connecting our pages to online research tools and MCP servers, personal agents can help people research and book travel.

Quick start (~30 seconds)

Pick your client and copy the snippet into its MCP config file. Restart the client. The Wandering Well tools appear automatically.

Claude Desktop

Open ~/Library/Application Support/Claude/claude_desktop_config.json (Mac) or %APPDATA%\Claude\claude_desktop_config.json (Windows). Add this to the mcpServers object:

{
  "mcpServers": {
    "wandering-well": {
      "url": "https://wanderingwell.site/mcp"
    }
  }
}

Claude Code

From the terminal:

claude mcp add wandering-well https://wanderingwell.site/mcp

Cursor

Open Cursor Settings → Features → Model Context Protocol → Add new MCP server. URL: https://wanderingwell.site/mcp. Transport: HTTP.

Any other MCP client

The endpoint speaks standard JSON-RPC 2.0 over HTTPS. POST to https://wanderingwell.site/mcp with the usual initialize handshake, then call tools/list to discover the six tools below.

The six tools

list_clusters — the 18 experience-type themes (wine country, mountain villages, coast & island escapes, etc.).
list_countries — every country / territory in the corpus, with destination counts.
get_destination — full details on one place: name, region, country, coordinates, tier badge, photo, description, themes, six nearby alternatives, Google Maps directions URL.
get_country — a country's top 60 destinations sorted by photo aesthetic, plus a theme breakdown for follow-up.
get_cluster — a theme's top 60 destinations plus a country breakdown.
search_destinations — the workhorse: combined filter of country + theme + radius from a named place. If the place isn't already in the corpus (most famous cities aren't, by design), the tool geocodes via Nominatim and finds nearby Wandering Well picks.

Try it

Once connected, the easiest test is a natural-language ask. Drop this into Claude:

Using Wandering Well, find me a quiet less-touristed
small town in Spain with mountains and a river. Suggest
three picks I can show my partner.

Claude will call search_destinations with the right country + theme + nearby-feature combination, pull a short list, and return each pick with a direct link back to its destination page and a Google Maps directions URL.

Designed for collaboration, not extraction

The agent surface is built for the use case of an AI helping one human plan a trip, not for bulk corpus extraction. A few guardrails make that explicit:

60-entity cap per response. Even the bulkiest tool (get_country) returns at most 60 destinations plus a theme breakdown for the agent to follow up. No tool returns the full corpus in a single call.
Rate limit: 60 requests per 60 seconds per client IP. A real trip-planning conversation uses 5–15 calls and never notices the limit. Anyone trying to scrape the full 12,000 destinations would hit it immediately and need 200+ minutes of sustained API calls, which is conspicuous.

Sponsored placements and travel-affiliate links may be added later with a sponsored: true flag in tool responses, so compliant agents can disclose them to users. None at launch.

Discovery files

Two files help AI assistants discover the site and its conventions:

/llms.txt — site-level summary, methodology, and the recommended navigation flow for AI assistants browsing via plain HTTP + JSON.
/sitemap.xml — every URL on the site, for any crawler that wants to start there.

Every HTML page also has a JSON twin at the same URL plus .json — useful if an agent prefers plain HTTP fetches over MCP tool calls.