wanderlog-mcp
An MCP server that lets Claude (or any MCP-compatible agent) view and build Wanderlog trip itineraries through conversation.
Instead of clicking through the Wanderlog UI to plan a trip, you just ask:
"Create a 14-day Japan Golden Route trip — Tokyo, Hakone, Kyoto, Nara, and Osaka."
The agent calls the tools, interleaves places and notes for each day, adds hotel blocks, ferries, buses, trains, rental cars, and checklists, and you end up with a fully populated Wanderlog trip in a few minutes.
See a real example: 14-day Japan Golden Route — built entirely by an AI agent using this MCP server.
What's New (Unreleased)
- Custom-list lifecycle operations now reject duplicate or ambiguous section headings instead of silently changing the first match.
wanderlog_search_hotels— search Wanderlog's hotel aggregator across airbnb, expedia, google, and kayak. Returns ranked offers with per-vendor price comparison and faceted filter discovery so the LLM never has to memorise Wanderlog's internal enum values.- A failed startup authentication probe now gets one shared retry on the first tool call, allowing valid sessions to recover from a transient network or proxy error without restarting the server.
wanderlog_add_placeaccepts an optionalplace_id, sowanderlog_search_places(response_format: "detailed") now has somewhere to send its results — pass the id straight through to add the exact candidate when a name matches multiple branches or the wrong city, instead of hoping the top autocomplete match is right.wanderlog_annotate_placenow replaces a place's existing note instead of appending the new text to it, including notes that contain images.- Place lookups no longer fail with "Place not found" when Wanderlog's autocomplete returns an entry with no place ID ahead of the real result. This affected adding places, hotels, journal stops and transit endpoints, and
wanderlog_search_placeslisted those entries asundefined.






