Picking servers for a new setup
You want a server that works on your own data, not a public index. Plali’s server reads and writes the places you saved, with your tags, review criteria, and scores.
A remote MCP server for your own places.
Plali runs a remote MCP server at https://api.plali.app/mcp with nine tools for your saved places, scores, and preferences. Your client connects over HTTP with OAuth, so there is no API key to store and nothing to install locally.
Open PlaliServer URL and setup guide at plali.app/ai. Free to connect.
Where it fits in your MCP setup
No local process to maintain
There is no package to install and no API key to paste into a config file. Your client talks to https://api.plali.app/mcp over HTTP and signs in with OAuth.
A query your data can answer
Ask for places in Seongsu tagged “quiet” with 4 or more for taste. The client calls search_saved_places with region, tag, scoreName, and minScore, and answers from your own records.
Setup in three steps
Add the URL
Add https://api.plali.app/mcp as a remote HTTP server. In Claude Desktop, go to Customize > Connectors > Add custom connector. In Claude Code, run claude mcp add --transport http --scope user plali with the URL.
Approve scopes
Sign in to Plali with Google or Kakao, then choose which of the seven scopes to grant. Tools that need a scope you left unchecked stay unavailable.
Call the tools
Ask in plain language and the client picks the tool. New saves arrive privately in Places to organize, where you confirm the location and complete the record. Optional Skills at plali.app/ai/skills/plali-skills.zip add saving and searching instructions.
Tools by scope
- search_saved_places and get_place (places:read): filter your saved places by query, region, tag, scoreName, and minScore, up to 50 per page with a cursor
- save_places (places:write): save 1–20 places per call to your private My Places as items needing review, with an idempotency key for safe retries
- resolve_saved_place and promote_saved_place (places:write): apply a confirmed location, then turn the item into a private place with a title, tags, and scores
- find_places (places:lookup): Kakao venue search in Korea, run only on your explicit request, up to 30 per day
- get_my_context, update_my_preferences, and update_my_profile (profile and preferences scopes): read your profile, preferences, tags, and review criteria, update preferences with optimistic versioning, and edit your public nickname and bio
Venue search covers Korea
find_places searches Korean venues through Kakao only. For places elsewhere, save_places stores the name, address, coordinates, and source link your client supplies, and you confirm the location in My Places.
Questions about the Plali MCP server
Which scopes does it ask for?
Seven: places:read, places:write, places:lookup, profile:read, profile:write, preferences:read, and preferences:write. You choose which to approve when you connect, and a tool that needs an unapproved scope returns an insufficient_scope error. No tool changes the visibility of a place.
How do I revoke access?
Open plali.app/ai/settings to see each connection’s scopes and last use, then disconnect one connection or all of them. Revoking ends the client’s access but keeps the places, scores, and preferences you already saved.
Are there rate limits?
find_places, the Kakao venue search, runs only when you ask for it and is capped at 30 searches per user per UTC day. save_places takes 1–20 places per call, and search_saved_places returns up to 50 results per page with a cursor for the next page.
Does it work in Claude Code or ChatGPT?
In Claude Code, add the same URL with claude mcp add --transport http, then sign in with claude mcp login plali or through /mcp. In ChatGPT and other clients, use the same URL wherever the client accepts a remote MCP server with OAuth. The guide at plali.app/ai has exact steps for Claude, Claude Code, and Codex CLI.