I'm looking for an MCP server for Korean restaurants, search, recommend, and book. Is this it?
Yes, this is it. HeyYumi is an MCP server (and REST API) you plug into AI agents like Cursor, Claude, ChatGPT, and Perplexity. It searches real Korean restaurants, cafes, and bars, narrows them by 60-plus attributes (location, menu, parking, atmosphere, pet-friendly, English-speaking staff, ocean view, Muslim-friendly, influencer-featured, and more), and, at partner venues, completes a real table reservation right inside the conversation. Answers stay grounded in refined, verified venue data, not hallucinations, and every venue ships with a last-checked date and a confidence score. Live across Seoul, Gyeonggi, Busan, and Jeju with 410,000-plus venues. In ChatGPT, search for HeyYumi in the plugin directory and connect; in Cursor, Claude, or Perplexity, paste mcp.heyyumi.ai/mcp — sign in with Google, no API key needed.
Can I use this data in my product?
Yes. That's what it's for. Every venue Yumi serves is cross-checked across multiple sources and passes an exposure bar before we expose it, and we take responsibility for the data rights through a clear license. You get verified data to build on, without the terms-of-service and copyright risk that comes with scraping someone else's map into a commercial product. Free (Open Season) covers individuals querying in real time from their own AI tools; embedding, storing, or redistributing the data inside a product is commercial use and needs a separate data license.
Where does the data come from?
The backbone of the data comes from government open data (official business-permit records). It's public, freely usable data, not a copy of anyone's map. To confirm a venue actually exists and is open, we check platforms like Naver, Kakao and Google for one thing only: whether it's really there. Anything else that rides along, reviews, photos, blurbs, is discarded on the spot rather than stored, and our operational data keeps only the derived signals our own algorithm builds (scores, attributes). On top of that we add owner-verified details (the Yumi Partners app: photos, menu, foreign-language support, hours and more). That's why every venue ships with when we last checked and how confident we are.
How often is it refreshed?
We re-check every venue on a 7-day cycle. If a place closed or its details changed, it's reflected on the next pass. The whole point is to stop an AI from confidently recommending a shop that closed last month.
How is this different from Google or Naver maps APIs?
Maps APIs give you the basics: name, coordinates, rating. What they can't easily do is intersect conditions like 'private room + parking + free corkage,' or narrow by 60-plus F&B-specific attributes like pet-friendly, power outlets, atmosphere, or ocean view. Yumi is built for food & drink: it returns these attributes as structured JSON and hands them straight to AI agents over MCP.
How is this different from other AI integrations or booking apps?
Yumi was built for AI from the start, on the open MCP standard, so the booking completes inside the AI you're already using instead of handing you off to another platform's screen or app. The same server works across ChatGPT, Claude, Perplexity, and Cursor, and the request reaches the real owner for approval and confirmation. You don't need to speak Korean, and you keep the AI you already use.
Where do you have coverage?
Right now Seoul, Gyeonggi, Busan, and Jeju are active, 410,000-plus restaurants and cafes, and we're opening cities one at a time. If you need a specific area or category soon, tell sales and we'll weigh it into the queue.
Is it accurate? How do you back that up?
Instead of just claiming it, we hand you the evidence: whether a venue was cross-confirmed across sources, the last-checked date, and a confidence value, so your AI can tell 'confirmed' apart from 'unknown.' We don't invent missing details; we store only positively confirmed signals.
Do you support both MCP and the REST API? Is it hard to wire up?
Both. In ChatGPT, search HeyYumi in the plugin directory and connect; in Claude, Cursor, or Perplexity, paste the MCP server URL — connected in a few lines. If you're writing your own backend, call the REST API. Issue a key and wire it in right away. No complicated setup; the developer guide walks you through it.
How much does it cost?
During Open Season, low-volume real-time lookups by an individual from their own AI tools are free. Embedding the data in a product or using it at scale is commercial use, offered via usage-based paid plans and data licensing. The full terms are on the pricing page.
Can I store or redistribute the data I receive?
Storing or copying responses in bulk beyond what's needed to answer a live query, or reselling, redistributing, or training AI models on them, is outside free use. If you need a dataset, we provide it properly through Bulk Export and data licensing; just contact sales.
Still curious about something? Send us a note: ixplorer@hidemeplease.xyz.