Sirv E-commerce Workflows

Product imagery operations from processing to publishing: AI Studio processes images, Sirv stores and delivers them, and the MCP tools below push results into Shopify, run saved pipelines, and enforce marketplace/supplier standards.

Prerequisite: the sirv-ai MCP server (setup in ../sirv-ai-studio/SKILL.md). Processing tools themselves (remove background, upscale, lifestyle, alt text) are documented there; this skill covers the store-facing layer.

Default Workflow

1. Check the connection state before promising anything: sirv_shopify_status returns connected, valid, and needsReconnect - if not connected/valid, stop and have the user connect the store in Sirv AI Studio.
2. Get products into the PIM when working catalog-wide: sirv_sync_shopify_products imports products and images from Shopify into Sirv AI Studio (inbound only; max 500 per call, default 100).
3. Process images per ../sirv-ai-studio/SKILL.md (single tools) or run a saved pipeline with sirv_execute_workflow for repeatable multi-step jobs.
4. Validate against marketplace rules before publishing: image review presets (Amazon, eBay, Shopify, Walmart) and sirv_run_supplier_autofix for automated correction.
5. Push results to the store with sirv_push_image_to_shopify - choose the upload_strategy deliberately (see below).
6. Verify in the store: image visible on the product, correct position/featured state, alt text present.

Shopify Tools

All accept optional orgId (UUID) - omitted means the default organization.

Push semantics - read before pushing

• productId accepts numeric or gid://shopify/Product/... form.
• upload_strategy:
  • add (default) - appends to the product's images.
  • replace_source - deletes one existing image (source_image_id required) then adds. Use for "replace this photo with the processed version".
  • replace_all - deletes every image on the product before adding. Destructive; confirm with the user before using it.
• set_as_featured: true puts the new image at position 1.
• Push attaches media at the product level, not to variants - it will not create or link variant images.
• Sync direction is inbound-only via MCP; pushing is the only outbound operation. Don't tell users "sync will upload your images to Shopify" - it won't.

Saved Workflows (Pipelines)

Users build pipelines visually in Sirv AI Studio (e.g. Remove BG → Validate → Autofix → Alt Text); agents execute them:

Execution facts: runs take 30 seconds to several minutes; every step spends credits per image (result reports totalCreditsUsed and per-step creditsUsed); interactive steps (review/imageReview) and client-side steps (saveToCsv) are skipped in MCP runs; saveToSirv, addMetaTags, and setProductMeta steps require the user's Sirv account to be connected. Results are per-image with success/partial/error status and finalOutputs URLs.

Prefer a saved workflow over hand-chaining single tools when the user will repeat the process - it keeps settings consistent across batches.

Store-facing step types worth knowing when reading a workflow: pushToShopify, channelReadinessGate (marketplace channel readiness), the DAM/PIM steps (saveToDam, linkToProduct, writeProductAttributes, addToCollection), and e-commerce AI steps that exist only as workflow steps, not standalone tools: smart crop, add shadow, color variants, ghost mannequin, image review. Workflow Triggers can run pipelines automatically on events (configured in the web app).

Supplier Portal (Supplier Portal Pro tier)

For teams receiving product images from suppliers/vendors:

Autofix spends AI credits per iteration. Fetch requirements first, pass the actual failed_rules in, and report appliedFixes from the response rather than assuming what changed.

Marketplace Compliance

• Image Review validates dimensions, background, watermarks, and frame fill against marketplace presets (Amazon, eBay, Shopify, Walmart) - run it before pushing to a marketplace channel, then autofix and re-review.
• Typical Amazon main-image loop: remove background (pure white) → validate → autofix failures → alt text → push.
• Delivery formatting (exact pixel sizes, padding, white canvas) belongs in Sirv Dynamic Imaging profiles (../sirv-dynamic-imaging/SKILL.md), not in repeated AI processing.

Red Flags

• Using replace_all without explicit user confirmation - it wipes the product's existing gallery.
• Expecting sirv_sync_shopify_products to upload images to Shopify (it imports, never exports).
• Pushing before checking sirv_shopify_status - a stale token fails with a confusing error; needsReconnect: true means the user must re-auth in the web app.
• Running sirv_execute_workflow on 100 images without checking credits (sirv_get_usage) or warning about runtime.
• Expecting variant-level image assignment from the push tool.
• Calling supplier tools for accounts without Supplier Portal Pro - they will fail on tier access.

Verification

• After push: fetch the product (store admin or sirv_list_products_with_assets) and confirm the image, its position, and alt text. PIM/DAM tool parameters live in ../sirv-ai-studio/references/asset-product-tools.md — note product deletes are permanent there, unlike soft asset deletes.
• After sync: report products.created/updated and images.imported/skipped from the result, not assumptions.
• After a workflow run: check per-image statuses; surface partial/error items with their step errors instead of declaring success on the batch.
• After autofix: re-run the review/requirements check to confirm the image now passes.