feat: add DoorDash connector

[letonchanh]

Summary

Exports your DoorDash order history including restaurant names, items ordered, dates, totals, and delivery status.

Scopes

• doordash.orders: Complete order history with restaurant names, items ordered, dates, totals, and delivery status

Build report

DoorDash Connector Report

Date: 2026-03-12
Agent: Claude Opus 4.6
Connector: doordash/doordash-playwright.js
Iterations: 1 (worked on first run)

Decision log

• Started by researching existing DoorDash scrapers (tshddx/doordash-scraper, theodoremoreland/OrderDashboard) to understand available data patterns. Both rely on session cookies; one uses batch API calls, the other does DOM scraping from the order history page.
• DoorDash uses Cloudflare bot protection, so automated login is unreliable. Chose manual browser login via showBrowser/promptUser as the primary auth path, with automated requestInput-based login as Tier 2 and env-var credentials as Tier 1. This supports email/password, Google Sign-In, and Apple Sign-In without any code changes.
• For data extraction, used network capture (captureNetwork) targeting the getConsumerOrdersWithDetails GraphQL query that DoorDash fires when loading /orders. This returns structured JSON with order IDs, store names, dates, totals, items, and status.
• Pagination is handled by scrolling to trigger additional GraphQL requests, with deduplication by order ID.
• For order details, navigated to each /orders/{uuid} page and captured getConsumerOrder GraphQL responses to extract item-level data (names, quantities, prices) plus delivery address and status.
• DOM scraping as fallback for both the order list and detail pages, in case GraphQL capture returns nothing (e.g., if DoorDash changes their API naming).

Surprises

• DoorDash's order history page uses infinite scroll, not traditional pagination — had to capture network requests triggered by scrollTo(document.body.scrollHeight) rather than clicking "next page" buttons.
• DoorDash login is multi-step: email field first, then a "Continue" button, then the password field appears on a second step. Standard single-form login code doesn't work.
• The account has orders in both USD and CAD (user was in Canada for some orders), so the total field preserves the original currency string rather than converting.
• Browser profile session reuse worked perfectly — after the first manual login, subsequent runs skipped login entirely.

Friction

• The validate.cjs --contribute script expects to be run from within a git repo. Had to clone the data-connectors repo and copy files into it before contributing. Would be nice if the script handled this automatically.

What worked well

• Network capture (captureNetwork) was the right approach — DoorDash's GraphQL responses contain clean, structured data with all the fields needed. No need to fall back to DOM scraping.
• Browser profile persistence saved time on subsequent runs — session cookies from the first login carried over.
• Three-tier login (session → automated → manual browser) meant the connector worked headlessly on re-runs after the initial manual login.
• Validator caught everything needed and the schema generator produced a good skeleton that just needed description enrichment.

---

Generated by the vana-connect skill.