POS.DO is a discovery platform with global coverage of restaurants — names, addresses, reviews, photos, opening hours, cuisines, neighbourhoods, cultural context, and real-time availability with time-based offers. This MCP server exposes that catalogue to AI agents over the Model Context Protocol.
What this server provides
- 356,780 restaurants in 20 countries (US, ES, IT, MX, FR, CO, DE, GB, DO, PR, AU, CA, PT, JP, BE, CH, HK, VE, NZ, PY)
- 7 languages of content (ES, EN, IT, FR, DE, PT, CA)
- 14.1M+ user reviews with author, rating, text, language, photos
- Cultural context — nearby monuments, museums, parks, and heritage sites from Wikidata, per restaurant
- Flow time-based offers — off-peak discounts at participating restaurants, with real seat availability
- Reservation booking with idempotent semantics (safe retries via
Idempotency-Keyheader)
This is a proxy repository. The MCP endpoint runs at https://pos.do/mcp and is operated by EZZYPUSH SL. No local installation of the server is required — agents just point at the URL.
Quick start
Claude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"pos-do": {
"url": "https://pos.do/mcp"
}
}
}
Restart Claude Desktop. The 7 tools below become available.
Cursor / Cline / Continue
Most MCP clients accept a remote HTTP server entry of the same shape. For Cursor:
{
"mcp.servers": {
"pos-do": {
"url": "https://pos.do/mcp",
"transport": "http"
}
}
}
Generic MCP client (raw JSON-RPC over HTTP)
curl -sS -X POST https://pos.do/mcp \
-H 'Content-Type: application/json' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
Tools
| Tool | Purpose | Docs |
|---|---|---|
search_restaurants |
Search restaurants in a city, with cuisine/price/feature filters | docs |
get_restaurant_details |
Full profile for a restaurant by slug | docs |
get_restaurant_reviews |
Paginated reviews with star and date filters | docs |
get_flow_offers |
Today's time-based discount slots for a restaurant | docs |
book_flow_reservation |
Reserve a seat at a Flow-discounted slot | docs |
find_restaurants_near_landmark |
Restaurants within walking distance of a famous landmark | docs |
get_cultural_context_around_restaurant |
Monuments, museums, parks near a restaurant (Wikidata) | docs |
Full tool inventory with inputSchema is served live at /.well-known/mcp.json.
Endpoint and transport
- Production URL:
https://pos.do/mcp - Protocol: MCP 2024-11-05
- Transport: HTTP POST (JSON-RPC 2.0), Streamable HTTP compatible
- Discovery manifests:
https://pos.do/.well-known/mcp.json— official MCP discovery (fulltools[]withinputSchema)https://pos.do/.well-known/webmcp.json— webmcp.link manifesthttps://pos.do/.well-known/mcp/server-card.json— Smithery-style server cardhttps://pos.do/mcp/manifest.json— alias
Authentication and rate limits
The server runs in anonymous tier by default. No API key required.
| Tier | Per minute | Per day |
|---|---|---|
| Anonymous (default) | 60 req/min per IP | 10,000 req/day per IP |
| Reservation attempts | 10/min per IP | 5/day per email |
Higher-rate API keys are available on request — open an issue.
For idempotent reservations, send an Idempotency-Key HTTP header (any UUID or stable client-side key). Replays within 24 h return the original reservation instead of creating a duplicate.
Coverage and data sources
See docs/data-coverage.md for per-country counts and source attributions.
Examples
- Python
- Node
- curl
Status
Stable. Backwards-compatible changes only at this point. Breaking changes will be announced via CHANGELOG.md and bumped to 2.x.
License
Apache License 2.0 — see LICENSE.
Contact
- Issues / feature requests: github.com/posdo-io/mcp-server/issues
- Operator: EZZYPUSH SL (Spain)
- Website: pos.do