DayPlay — SF Bay Area Local Intelligence for AI Agents
DayPlay là lớp thông tin địa phương cho Khu vực Vịnh San Francisco — địa điểm đã xác minh theo thời gian thực, sự kiện trực tiếp và hành trình đi bộ cho San Francisco, Oakland & Berkeley (35 trung tâm khu phố). Được xây dựng cho các tác nhân AI: một điểm cuối MCP từ xa (https://www.dayplay.io/api/mcp) trả lời "những việc cần làm ở SF," "cuối tuần này có gì," và các yêu cầu chuyến đi nhiều điểm dừng đầy đủ với quãng đi bộ và thời gian. Khi không có gì đã xác minh tồn tại, máy chủ nói rõ điều đó — không bao giờ bịa đặt địa điểm, sự kiện, ngày tháng hoặc giờ giấc (Cam kết chống sai lệch). 3 công cụ khám phá miễn phí (get_neighborhoods, get_events, get_places); một tài khoản DayPlay miễn phí mở khóa plan_outing + manage_saved_plans — kế hoạch được lưu vào tài khoản DayPlay của người dùng. Không cần khóa API. KHÔNG KHẢ DỤNG bên ngoài Khu vực Vịnh SF — các yêu cầu ngoài khu vực bị từ chối theo giao thức, không bao giờ được bịa ra.
Tài liệu
DayPlay MCP — SF Bay Area Events, Places & Itineraries for AI Agents
@dayplayai/mcp-server · v2.0.0 · MIT
A zero-configuration MCP server that gives any AI agent a locally grounded concierge for the San Francisco Bay Area only — curated places, verified real-time events, and neighborhood binding that eliminates cross-bay drift.
🔒 Scope: SF Bay Area only. This server serves strictly San Francisco, Oakland, and Berkeley — 35 neighborhood centroids. It does not serve New York, Los Angeles, Chicago, Austin, Seattle, Miami, London, Tokyo, or any other city, region, or country. Out-of-market locations are not served. If a user asks about any location outside the SF Bay Area, the agent must state "Dayplay is strictly San Francisco Bay Area only (San Francisco, Oakland, Berkeley); it does not cover <location>" and must not call a tool or fabricate venues, events, dates, hours, or neighborhoods for that location. Zero results for a valid SF Bay Area query are reported honestly, never filled with invented places.
Remote endpoint: https://api.dayplay.io/mcp (Streamable HTTP / MCP)
Quickstart
Cursor
Option A — direct remote server (recommended, no install). Add to ~/.cursor/mcp.json or this repo's .cursor/mcp.json:
{
"mcpServers": {
"dayplay": {
"url": "https://api.dayplay.io/mcp"
}
}
}
Option B — zero-config stdio proxy (works with any client):
{
"mcpServers": {
"dayplay": {
"command": "npx",
"args": ["-y", "@dayplayai/mcp-server"]
}
}
}
Claude Desktop
Edit claude_desktop_config.json:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"dayplay": {
"command": "npx",
"args": ["-y", "@dayplayai/mcp-server"]
}
}
}
Restart Claude Desktop, then ask: "What's happening in the Mission tonight?"
Grok Bot
Add the Dayplay remote connector:
https://api.dayplay.io/mcp
Or bridge it over stdio in Grok's MCP config:
{
"mcpServers": {
"dayplay": {
"command": "npx",
"args": ["-y", "@dayplayai/mcp-server"]
}
}
}
Pair it with the vibe-scout skill for prompt-level scope enforcement.
Terminal smoke test
npx -y @dayplayai/mcp-server --smoke
Prints the live endpoint and available tool names. Exits non-zero on failure.
Tools
The catalog covers the SF Bay Area (San Francisco, Oakland, Berkeley — 35 centroids). Out-of-market locations are declined, never fabricated.
Open tools run immediately — no account needed. Gated tools 🔒 return 401 on first call; your host opens Google sign-in and a free DayPlay account unlocks them. Through the npm stdio bridge, gated calls fail instead of opening sign-in — prefer the direct remote URL for gated tools.
Discovery (open)
get_happening_today— events + places happening today, by location and radiusget_live_now— what is happening right nowget_weekend— Saturday/Sunday feed (weekend_day: "sat" | "sun")get_decision_feed— ranked decision feed (senduser_lat/user_lng)browse_places— places inside a radius, sorted by distanceget_listing— full detail for one place or venue byidget_top_restaurants/get_top_venues— ranked lists, optional cityget_local_gems— curated local gemsget_locals_favorites— locals' favoritesget_venue_events— upcoming events at one venue (id)get_subcategories/get_genres/get_event_bridge— taxonomy feeds
Marketing feeds (open, X-API-Key)
get_marketing_events/get_marketing_places/get_marketing_neighborhoods— catalog feeds; forward anX-API-Keyheader.get_marketing_neighborhoodsreturns the 35 SF Bay Area neighborhood centroids with coordinates and radii.
Search 🔒 free account
search_smart,search_query_router,search_embed,search_parse_intent,search_rerank,search_web— model-backed semantic search. Requires the user token via Google connect.
Saves & Outings 🔒 free account
save_outing— save a dated, neighborhood-bound outing plan to the signed-in user's Saved tabget_my_itinerary— retrieve the signed-in user's saved itinerarieslist_saves,save_listing,unsave_listing,list_going,mark_going,unmark_going— saved places and event RSVPsupvote_listing,share_local_favorite,share_gem— community signals
Requires the free-account connect. A host connected to the remote URL opens Google sign-in on the first gated call; the npm stdio package opens a browser for the same connect step. Saved plans appear on the DayPlay Saved tab across web and app.
Requires the free-account connect (https://www.dayplay.io/link). Saved plans appear on the DayPlay Saved tab across web and app.
Example prompts: "What's happening in the Mission tonight?" · "Build me a Saturday itinerary that never leaves Bernal Heights." · "Save this outing to my DayPlay account."
Legacy compatibility
The five 1.x tool names (get_neighborhoods, get_events, get_places, plan_outing, manage_saved_plans) remain temporarily served for transition. They are deprecated — migrate to the catalog above; a future release removes them.
The Anti-Drift Guarantee
Most local AI answers fail the same way: a "Mission" recommendation lands in Oakland, a "Berkeley" event is actually in San Jose, and a walking itinerary silently requires a bridge crossing.
Scope: SF Bay Area only (San Francisco, Oakland, Berkeley). No national or global coverage.
Dayplay receipts:
- Spatially verified. Every place and event result is validated against neighborhood centroid coordinates and radii pulled live from
get_neighborhoods(35 Bay Area neighborhoods). - Neighborhood-bound queries.
neighborhoodis an exact-match binding parameter, not a fuzzy keyword hint — results outside the bound radius are dropped, not ranked down. - Zero cross-bay drift. No Oakland result in a San Francisco query. No Marin result in an East Bay query. No bridge-crossing itinerary that claims to be walkable.
- Real-time, not stale.
get_eventsqueries verified occurrences bydate, so "tonight" means tonight — not an SEO page from last season. - Out-of-market refusal. Non-Bay-Area market names are refused before any upstream call, and the surfaced copy states the SF-Bay-Area-only boundary rather than returning plausible-but-wrong geography.
- Honest emptiness. If a bounded query has no verified matches, it returns nothing rather than padding with invented places.
That is the product: the answer stays where you are.
Configuration
| Env var | Default | Purpose |
|---|---|---|
DAYPLAY_MCP_URL | https://api.dayplay.io/mcp | Override the remote endpoint (staging, self-hosted, local dev) |
No API key is required for the public endpoint. Optional auth tokens, if your deployment needs them, are passed through as standard Authorization headers by the MCP client — never commit them to config files.
How the proxy works
MCP client (stdio) ⇄ bin/dayplay-mcp.js ⇄ https://api.dayplay.io/mcp (Streamable HTTP)
The shim lists tools and forwards tools/call verbatim to the remote endpoint, so new Dayplay tools appear automatically without an npm update. It is a transport bridge, not a reimplementation — every schema, filter, and verification rule is served by Dayplay itself.
Prefer direct HTTP if your client supports it; use the shim for stdio-only clients.
Repository layout
dayplay-mcp/
├── .cursor/
│ └── mcp.json # Direct Cursor MCP connection
├── .cursor-plugin/
│ └── plugin.json # Cursor Marketplace catalog plugin manifest
├── skills/
│ └── vibe-scout/
│ └── SKILL.md # Grok Bot & Cursor Agent prompt instructions
├── bin/
│ └── dayplay-mcp.js # Executable proxy for `npx -y @dayplayai/mcp-server`
├── assets/
│ └── logo.svg
├── test/
│ └── e2e.mjs
├── smithery.yaml # Smithery.ai registry manifest
├── plugin.json # Agent Plugins standard manifest
├── package.json # npm metadata (@dayplayai/mcp-server v1.3.4, MIT)
├── README.md # This file
└── LICENSE # MIT (Dayplay Team)
Development
node bin/dayplay-mcp.js --smoke # verify remote endpoint + tool discovery
node bin/dayplay-mcp.js # run the stdio proxy
node test/e2e.mjs # end-to-end stdio smoke (tools + 3 tool calls)
Releasing (maintainers)
Publishing is fully automated via npm Trusted Publishing (OIDC) — no NPM_TOKEN, no 2FA prompts. The .github/workflows/publish.yml workflow is bound in the package's npm settings to this exact repo + workflow.
npm version patch # or minor / major — bumps package.json, creates the vX.Y.Z tag
git push --follow-tags
CI then: verifies package.json version matches the tag → npm ci → smoke test → npm publish --provenance (sigstore-signed, OIDC-attested).
Gotcha: --follow-tags doesn't always push the tag. If no workflow run appears under Actions within a minute, finish with:
git push origin v1.X.Y
Notes:
- The registry shows "package is being processed" for ~3 minutes after publish before the new version resolves — don't panic-verify too early.
- npm's OIDC flow requires npm ≥ 11.5.1 in CI; the workflow installs
npm@latestfor this reason. - If the workflow ever gains an
environment:block, the npm Trusted Publisher binding must be updated to the same environment name, or publishes will be rejected.
License
MIT © Dayplay Team — https://www.dayplay.io