Restaurant finder

Restaurant Finder

Discover restaurants, scan availability fast, book interactively.

Config: Load config.md for personal details (name, email, phone for booking forms). Platform reference: See references/booking-platforms.md for per-platform form-filling notes and Chrome MCP extraction scripts.

Connectors

Phase	Tool	Why
Discover restaurants	Rube (Google Maps)	Reliable ratings, location-aware
Scan availability	Firecrawl	Parallel across all date × restaurant combos
Scan availability (fallback)	Chrome MCP	When Firecrawl credits exhausted — sequential but free
Book	Chrome MCP	Interactive — click slots, handle seating, fill forms, confirm

Workflow

Step 1: Gather Details

Parse from the user's message. Only ask if critical info is missing:

Input	Required?	Default
Restaurant(s) OR location	Yes	—
Date(s)	Yes	—
Party size	Yes	2
Time	No	6:00–8:00 PM window
Cuisine	No	Any
Special requests	No	None

Two modes:

• Specific restaurants named → skip discovery, go straight to availability scan (Step 3)
• Location/vibe only → run discovery first (Step 2)

Step 2: Discover Restaurants (Skip if restaurants specified)

Use Rube to search Google Maps for top-rated restaurants near the location.

Search with: location, radius (~1 mile), type: restaurant, minimum rating 4.3, cuisine keyword if specified.

Two location anchors (from + to): Run both searches in parallel. Label results by anchor.

Present top 10 sorted by rating before proceeding:

Top-rated near [location]:

1. [Name] — [rating] [cuisine] [address]
2. ...

Which should I check for availability?

Fallback: If Google Maps isn't connected in Rube, use Firecrawl or Exa to search for restaurants, or Chrome MCP to search Google Maps manually.

Step 3: Scan Availability

Try Firecrawl first (fast, parallel). If credits exhausted, fall back to Chrome MCP (sequential but free).

Firecrawl Approach (Primary)

Fire all date × restaurant combinations in parallel — one scrape per combo, all in a single message.

Credit cost: Each scrape with waitFor costs ~5 Firecrawl credits. A 2-restaurant × 3-date search = 30 credits.

OpenTable search URL (works as both platform detection and availability scan):

https://www.opentable.com/s?covers={party_size}&dateTime={YYYY-MM-DD}T{HH:MM}&term={restaurant_name}+{city}

If the restaurant appears in results → it's on OpenTable. If not → try Resy/Tock.

Use JSON extraction with a schema that captures restaurant name, available times, and booking URL. Set waitFor: 5000 for JavaScript rendering.

Resy — SPA, direct restaurant URLs are unreliable. Scrape city listing page (resy.com/cities/{city-code}) in markdown format with waitFor: 5000 and search for the restaurant name.

Tock — direct pages usually render: exploretock.com/{restaurant-slug}

If Firecrawl returns empty: Increase waitFor to 8000–10000ms, try markdown format, or fall back to Chrome MCP.

Chrome MCP Fallback

Sequential (one page at a time) but reliable and free. Navigate to each OpenTable search URL and extract availability with evaluate_script. See references/booking-platforms.md for the extraction script.

Step 4: Present Results

# Reservations for [Party Size]
## [Date range] around [Target Time]

### [Restaurant Name]

| Date | Available Times | Notes |
|------|----------------|-------|
| **Fri 4/3** | 6:15 PM, 6:30 PM, 6:45 PM | On target |
| **Sat 4/4** | 9:00 PM, 9:15 PM | Late only |
| **Sun 4/5** | No availability | Fully booked |

### Bottom line
[Quick recommendation — best match to what the user asked for]

Sort: near-target availability first, then late/early, then fully booked, then phone-only.

Step 5: Book via Chrome MCP

Once the user selects a restaurant + date + time:

1. Navigate to the search results page (NOT direct restaurant URLs — some 404): https://www.opentable.com/s?covers={n}&dateTime={date}T{time}&term={restaurant}+{city}
2. Click the target time slot via evaluate_script
3. Handle seating options if they appear (bar, dining room, patio) — present choices and wait
4. Fill personal details from config.md using the React-aware value setter (see references/booking-platforms.md)
5. Opt out of marketing — uncheck email/SMS opt-ins
6. Confirm before submitting — show restaurant, date, time, party size, seating type. Only proceed after explicit approval.
7. Submit and capture confirmation number

Step 6: Save to Favorites (Optional)

After a successful booking, offer to save the restaurant for future trips.

Edge Cases

Scenario	Response
No availability at target time	Check ±30 min, then ±1 hr. If still nothing, ask to expand.
Large parties (6+)	Most platforms cap at 6 online. Note to call directly.
Restaurant not on any platform	Include with phone: "Not on online booking — call [phone]."
Firecrawl credits exhausted	Switch to Chrome MCP fallback. Note it'll be slower.
Auth modal during booking	Ask user to log in manually in the browser, then retry.

Do / Don't

Do:

• Fire all Firecrawl scrapes in parallel (one per date × restaurant combo)
• Use OpenTable search URLs — they double as platform detection AND availability scan
• Use evaluate_script over click/fill for Chrome MCP interactions
• Book from the search results page (click time slot → booking flow)
• Take snapshots before and after form filling
• Uncheck marketing opt-ins

Don't:

• Construct direct Resy restaurant URLs (SPA — they 404)
• Navigate to opentable.com/r/{slug} for booking (can 404 — use search page instead)
• Submit without explicit confirmation
• Skip snapshot verification before submitting

Customization Notes

• Discovery method — Rube (Google Maps via Composio) or Firecrawl/Exa search. The key requirement is rated restaurant search within a radius.
• Booking platforms — The skill covers OpenTable, Resy, and Tock. Add platforms by extending the references file.
• Form fill details — Edit config.md with your name, email, and phone for form autofill.
• Browser tool — Chrome DevTools MCP is used for booking. The AI assistant Chrome profile should be logged into your booking platforms.