Daiso MCP

What it does

Daiso MCP is a public remote server, run by an independent developer, for everyday look-ups in South Korea. You can search Daiso products and stores, check stock, and see where an item sits on a store's shelves. The same goes for products, stores and stock at GS25, 7-Eleven, CU, Emart24 and Olive Young. For Olive Young it can also pull the day's deals, priced as shown at the time of the call. It also finds nearby restaurants and cafes through Naver local search, reads fuel prices from Opinet, and shows films, showtimes and remaining seats for CGV, Megabox, Lotte Cinema and the art-house screens booked through Dtryx. Daiso Japan is covered too, for online products, yen prices and stores, but not store stock.

The tool list is fixed at 10 general tools. You pick the chain with a `service` argument and, where a chain has more than one action, an `operation` such as `by_name`, `price` or `seats`. The docs say the count will stay at 10. Olive Young's daily deals show how: they arrived later as `get_catalog` with operation `today_deals`, and the internal name `oliveyoung_get_today_deals` was never put on the public list. Of the 44 older call names, `compare_products` kept its name and the other 43, such as `daiso_search_products`, still answer as deprecated aliases that point to their replacement in `_meta.replacement`. `tools/list` no longer shows them. The endpoint root reports 15 services and 10 tools.

You connect to https://mcp.aka.page over Streamable HTTP. There is nothing to install and no key to give. It runs on Cloudflare Workers, and the maintainer holds the upstream keys. Stock checks need a product found by a search first, or the Daiso `by_name` operation. The same backend also serves a REST GET API, plain HTML pages under /browse/ (the starting page is /browse/quick) and an `npx daiso` CLI for clients that cannot speak MCP.

What it needs

An MCP client that can add a remote Streamable HTTP server, such as Claude, Claude Code or Home Assistant. Self-hosting means deploying the repository to Cloudflare Workers with your own Opinet, Naver, Kakao and Supabase keys (see CONTRIBUTING.md).

Limits

• Region Covers South Korean chains plus Daiso Japan. Answers come mostly in Korean.
• Not covered Daiso Japan has products, yen prices and stores only. Japanese store stock is not supported.
• Not covered Lotte Mart support has ended.
• Limit The old per-service tool names are deprecated aliases. They still work but are hidden from tools/list, so new setups should use the 10 current tools.
• Rate limit REST GET paths under /api/oliveyoung/, /api/cgv/, /api/cu/ and /api/gs25/ are capped at 3,000 calls per IP per day (KST). MCP POST calls are not counted by that cap.
• Rate limit Opinet runs on a free key budgeted at about 1,500 upstream calls a day and is cached for 20 to 60 minutes, so fuel prices can be a little stale.
• Limit Data is read from the chains' public sites, so a change on their side can break a service until the server is updated.
• Limit Stock and seat counts can change after a look-up. An unconfirmed count is reported apart from sold out.

Covers South Korean chains only, and answers come mostly in Korean. Lotte Mart support has ended. The public server caps the REST GET paths for Olive Young, CGV, CU and GS25 at 3,000 calls per IP per day (KST). In the code that cap is enforced only on GET requests under /api/oliveyoung/, /api/cgv/, /api/cu/ and /api/gs25/, not on MCP POST calls. Opinet data runs on a free key budgeted at about 1,500 upstream calls a day and is edge-cached for 20 to 60 minutes, so fuel prices can be slightly stale. Relayed upstream calls also have shared daily and per-minute budgets. Data is read from the chains' public sites, so a site change can break a tool until the server is updated.

Who it suits

Suits someone in Korea, or building for Korean users, who wants an assistant to check whether a nearby Daiso or convenience store has an item, find a cheap petrol station, or see cinema seats, with nothing to set up. Daiso Japan product and store search is a smaller extra. It reads public data and the only tool that stores anything is the request tool, and no sign-in is needed. Everything goes through one maintainer's hosted service and scraped public sites, so look elsewhere if you need guaranteed uptime or an official feed from these chains. Heavy users are told to deploy their own copy.