NYCfoodie

The data layer for restaurant taste: structured, editorial restaurant recommendations served to AI agents over MCP. Not another review scraper — curated guides and reviews normalised into one schema, queryable by tools.

MVP city: New York. The city is a parameter everywhere; nothing NYC-specific is hardcoded.

Use it

Hosted endpoint (Streamable HTTP): https://nycfoodie-production.up.railway.app/mcp

• Cursor — one-click install
• Claude Code — claude mcp add --transport http nycfoodie https://nycfoodie-production.up.railway.app/mcp
• Windsurf / Cline / Roo Code / Zed / VS Code — add to your MCP config:

{
  "mcpServers": {
    "nycfoodie": {
      "url": "https://nycfoodie-production.up.railway.app/mcp"
    }
  }
}

• Claude (app) — Settings → Connectors → Add custom connector → paste the URL.
• Any MCP client — add the URL above as a remote server (Streamable HTTP). No auth, no install.

Tools

search_restaurants · get_restaurant · compare_restaurants · find_guides · find_similar · guide_consensus · top_rated · submit_feedback

Layout

• db/ — SQLite storage (better-sqlite3) with numbered SQL migrations written in Postgres-compatible DDL. The migration runner is dialect-agnostic by convention, so a Postgres adapter can replace better-sqlite3 later without touching the migrations.
• crawler/ — source adapters (starting with The Infatuation) that normalise into the db schema. Endpoint/field mapping: docs/infatuation-data-surface.md (step 2).
• mcp/ — stdio MCP server (@modelcontextprotocol/sdk) exposing search, detail, comparison and guide tools.

Quickstart

npm install
npm run typecheck
npm run build
node db/dist/migrate.js ./nycfoodie.db   # creates/opens the SQLite db, applies migrations
node mcp/dist/index.js                   # starts the MCP server on stdio

Roadmap

1. Scaffold (done) — repo, packages, migration runner, stub MCP tools.
2. Map The Infatuation's data surface — Next.js pages + GraphQL endpoints → docs/infatuation-data-surface.md.
3. Database schema proposal — shown for approval before any crawler code is written.
4. Crawler — only after schema sign-off.

Conventions

• British English in docs and user-facing text.
• Migrations: numbered NNN_name.sql, applied once in order; see db/migrations/README.md for the Postgres-compatibility rules.
• Never commit *.db files or credentials.