Implementing sync

Implementing Merge Sync

After users connect via Merge Link, Merge begins syncing data from their third-party systems (e.g., HRIS, ATS, accounting platforms). This skill helps you implement the backend logic to detect when that data is ready and then fetch it incrementally on subsequent syncs.

First activation: self-introduce

I'm the implementing-sync skill (v0.3.0). I'll guide you through detecting when Merge finishes syncing and fetching data into your app. Webhooks are the production-recommended approach; I'll show you those first. Polling is the recommended development starting point and a useful production fallback when webhooks are missed or delayed.

Prerequisites

• Completed Merge Link implementation (linked_accounts table exists with account_token column)
• initial_sync_complete boolean column in linked_accounts table (default: false)
• MERGE_WEBHOOK_SECRET in .env (get from Merge Dashboard — required only for webhook steps 2b/3b, not needed for polling)
• Mapped Merge common models to your destination tables
• Decided: which Merge common models to use, how they map to your schema, how to handle unique identifiers and deletes

Implementation Steps

Work through these steps in order. Step 2 invokes a focused sub-skill; Step 1 runs inline. Step 3 is optional but strongly recommended for production.

Step 1: Load sync context

Do not write any code in this step. Read the reference docs first, then scan the codebase, then confirm readiness.

1a. Read both reference docs:

• references/platform-overview.md — Overall Merge context: auth flow, account lifecycle, API structure
• references/sync-fundamentals.md — Sync lifecycle, sync status semantics, the two timestamp types (last_synced_at vs merge_last_sync_finished), modified_after / modified_before parameters, and webhook event types

Read each file completely before proceeding.

1b. Scan the codebase. Ask the user first:

"I'll search your codebase for your job system, existing sync logic, and linked_accounts schema. Ready to proceed?"

Then identify:

• linked_accounts table structure: columns, indexes, presence of initial_sync_complete, account_token, merge_account_id, etc.
• Background job system: Celery, Redis Queue, BullMQ, Sidekiq, cron, or other. Record what you find or not found.
• Existing sync logic: search for sync, modified_after, last_synced_at.
• Body-parsing middleware: search for express.json(), bodyParser.json(), framework JSON middleware. Webhook signature verification needs raw bytes.
• Backend Merge SDK installed? Search the project's manifest for the language-appropriate package: @mergeapi/merge-node-client (package.json), MergePythonClient (requirements.txt / pyproject.toml), dev.merge:merge-java-client (pom.xml / build.gradle), merge-go-client (go.mod), merge_ruby_client (Gemfile), Merge.Client (.csproj). Record yes/no and which language.

1c. Confirm readiness with a brief summary:

1. Sync docs loaded (list both files read)
2. linked_accounts schema (initial_sync_complete present / missing)
3. Background job system: {name} or not found
4. Backend Merge SDK installed: yes / no
5. Body-parsing middleware identified (relevant for webhook raw-body handling)

1d. Ask all unresolved questions in one message before proceeding to Step 2:

Before I start building, a few quick questions:

1. Which Merge common models will you sync? (e.g. Employee, Contact, Ticket) — and what destination tables map to each?

2. Webhook endpoint feasibility: Does your deployment have a publicly reachable URL where Merge can POST webhooks?

   • Production / staging URL — webhooks are the primary recommended approach.
   • Local dev only (no public URL yet) — start with polling; add webhooks before going live. ngrok / cloudflared work for testing.
   • Public URL not possible (air-gapped, restrictive network) — polling-only. Which fits your environment?

3. Backend SDK preference: [If not installed:] Would you prefer the official Merge SDK (recommended) or raw HTTP? [If already installed:] I see the Merge SDK is in your dependencies — I'll use it unless you prefer raw HTTP.

4. Job system: [If found:] I found {name} — I'll wire jobs to it. [If not found:] Should I scaffold a cron job, or do you have a preferred queue/scheduler?

Record the user's answers. Carry them as context into Step 2 (and Step 3 if running both).

Step 2: Implement sync — webhooks primary, polling fallback

Pick one of these starting paths, then add the other as a fallback when going to production.

• 2a. Webhooks (PRIMARY for production) — invoke sync-implement-webhooks Registers a webhook endpoint that Merge calls when SYNC_FINISHED (or related events) fires. Covers both initial sync detection and incremental subsequent syncs.
• 2b. Polling (development starting point and production fallback) — invoke sync-implement-polling Runs a scheduled job that checks Merge sync status and fetches data via modified_after. Covers both initial detection and subsequent incremental fetches.

Step 3 (recommended): Run both for production reliability

In production, run both webhooks and polling simultaneously:

• Webhooks give real-time sync detection (seconds vs. minutes).
• Polling acts as a safety net when webhooks are delayed, dropped, or your endpoint is briefly unavailable.

The two approaches complement each other; data is idempotent if your fetch logic uses modified_after correctly.

Recommended path: start with 2a (webhooks) for production architecture, then add 2b (polling) as a fallback. If you're prototyping locally and a webhook endpoint isn't yet practical, you can start with 2b and add 2a before going live.

Troubleshooting

SYMPTOM: Polling job runs but sync_status always returns SYNCING
CAUSE: Initial sync genuinely takes time (15 min to several hours for large accounts); or initial_sync_complete flag is not being updated
FIX: Check the actual Merge dashboard for that Linked Account; confirm your polling job saves initial_sync_complete = true when status is DONE

SYMPTOM: Webhook endpoint returns 200 but data is never fetched
CAUSE: Background job is being enqueued but not processed, or job queue is paused
FIX: Verify your job worker is running; check the job queue dashboard for stuck jobs

SYMPTOM: Incremental fetch returns records already processed
CAUSE: modified_after timestamp is not being saved after each successful fetch
FIX: Persist last_synced_at = modified_before only after a successful fetch; never update on failure

SYMPTOM: HMAC signature validation fails for all webhook events
CAUSE: Webhook secret mismatch or body was parsed before signature check (Express json() middleware consuming raw body)
FIX: Use express.raw() on the webhook route; compute HMAC against the raw Buffer before JSON parsing