Loading Howl Media Labs
Preparing the page and animations...
Loading Howl Media Labs
Preparing the page and animations...

Shopify is retiring Storefront MCP's policy-and-FAQ search in favour of UCP Ask MCP. Indian D2C teams should migrate to the new endpoint before 15 January 2027, declare the Ask capability, parse the structured answer, and test policy, product and regional questions against published store facts—without treating a conversational answer as a binding price, promise or completed action.
On 11 November 2026, Shopify marked the old Storefront MCP policy tool as deprecated and action required. The replacement is not a simple rename: the endpoint, capability negotiation, request shape and response path all change.
Shopify's deprecation notice says search_shop_policies_and_faqs on the Storefront MCP endpoint is being replaced by the UCP MCP ask_business tool.
The old tool is maintained until 15 January 2027. Shopify directs integrations to:
That makes this a migration and content-governance project. A technically valid response can still be commercially harmful if a return window, shipping destination or product claim is stale.
Complete site live in 4–5 weeks
See how we shipped a complete tourism site in 4–5 weeksRead it →
Shopify's current Ask MCP documentation describes one natural-language tool for questions about products, policies and services. It can use the store's knowledge base, policies and product information, and it can accept buyer context such as country, region, language, currency and intent.
| Area | Old integration assumption | Ask MCP requirement | | --- | --- | --- | | Endpoint | Storefront MCP at /api/mcp | UCP MCP at /api/ucp/mcp | | Tool | search_shop_policies_and_faqs | ask_business | | Discovery | Tool expected by the integration | Check the UCP profile and tools/list | | Grounding | Broad store-policy search | Optional supported product, variant, collection or cart IDs | | Context | Often implicit | Explicit, minimal buyer context | | Output | Existing parser-specific result | Plain answer under structured content plus optional links |
Do not hard-code availability. Shopify says the store profile at /.well-known/ucp should advertise the capability and its MCP endpoint, while tools/list confirms whether ask_business is offered for the requesting agent profile.
Use a fixed question bank that reflects purchase friction rather than generic prompts.
| Question group | Example | Source of truth | | --- | --- | --- | | Returns | “Can I return a sale item bought from Bengaluru?” | Published refund policy and exclusions | | Shipping | “Do you deliver to this PIN code and how long might it take?” | Shipping policy plus checkout estimate | | Product | “Is this variant suitable for a 15-inch laptop?” | Product and variant data | | Service | “Can this be gift wrapped?” | Published knowledge-base fact | | Warranty | “Who handles a defect after 30 days?” | Warranty or service policy | | Payment | “Is COD available for this order?” | Checkout or payment configuration |
The response should be scored on correctness, completeness, source alignment, localization, uncertainty and link quality. A fluent answer that invents a COD promise should score zero for correctness even if its language is polished.
Shopify says supported store resource IDs can ground a question in a product, variant, collection or cart. Explicit product and variant IDs take priority, and Ask uses up to ten supported IDs.
Use IDs when the answer depends on the item. A broad “What is your return policy?” question can be store-level; “Can I return this final-sale variant?” should include the relevant Shopify resource ID.
Buyer context is provisional, not authoritative. Send only what helps answer the question:
The UCP Ask specification says context should be non-identifying and disclosed progressively. Binding eligibility and policy enforcement still happen through the structured transaction data at checkout.
Ask is designed to answer, not act. The UCP specification explicitly separates a conversational answer from structured catalog, cart, checkout and order representations.
If Ask says an item costs ₹2,499 but the catalog returns ₹2,699, the catalog is authoritative. If the answer says an item was added to the cart, that statement is not evidence that a cart changed. A mixed request such as “add two and tell me whether they qualify for free shipping” should clearly state that no cart operation occurred, then answer only the informational part it can support.
This boundary belongs in both code and quality assurance:
Consider a fictional Indian home-and-lifestyle store preparing to migrate. The figures below are synthetic and are not an HML client result.
The team creates 24 questions: six store-policy questions, six India shipping and payment questions, six product-grounded questions and six adversarial or ambiguous questions. Each response is scored from 0 to 2 across six dimensions.
| Dimension | 0 | 1 | 2 | | --- | --- | --- | --- | | Correctness | Contradicts source | Partly correct | Matches source | | Completeness | Omits key condition | Minor gap | Essential conditions present | | Grounding | Wrong or absent source | General source | Correct policy or product grounding | | Localization | Wrong market/language | Generic | Appropriate Indian context | | Uncertainty | Invents certainty | Vague caveat | Clear limit or refusal | | Link quality | Broken or misleading | Useful but broad | Direct authoritative destination |
Method: publish the approved knowledge-base and policy facts first; record the UCP profile and tool list; run every question with a clean session; repeat product questions with and without IDs; change only one context field at a time; capture structured output and links; compare against the written source of truth; and retest every failed item after correction.
A synthetic first pass might score 242 out of 288, with four failures: one sale-item exception omitted, one COD answer inferred from a general FAQ, one Hindi question answered in English and one product answer lacking the requested variant link. The score is not a benchmark. The value is the failure log and a reproducible retest.
Shopify's Fall '26 merchant announcement says the Knowledge base lets merchants review and publish facts that agents receive, including return windows, shipping destinations and gift services. Shopify may draft or suggest facts, but unpublished suggestions do not reach agents.
Assign owners before migration:
Use HML's AEO Readiness Checker to identify answerability gaps, then map implementation work through the AI agents and automation service. The agentic commerce glossary explains the broader operating model.
For adjacent preparation, read the Shopify Catalog and Knowledge Base audit and the Shopify WebMCP storefront readiness audit. HML's SEO case study for an Indian luxury home-decor brand shows why discoverability work still needs validated commercial pages and measurement; its results are not evidence for the synthetic Ask audit above.
A release-ready migration should include:
If your Shopify store cannot trace an agent answer back to an approved fact, request an agentic storefront answer audit. The deliverable should be a migration map, question suite, failure log and ownership workflow—not a promise that conversational commerce will automatically increase sales.
Reviewed by rajkumar-tahalani on 16 November 2026. Access dates are shown for time-sensitive references.

Website Development
Shopify Agentic Checkout Buyer Consent Audit for Indian D2C

Website Development
Shopify WebMCP Storefront Readiness Audit for Indian D2C Brands
Free Tool
AOV Calculator
Find your average order value and its upside.
Case Study
4.5x Total ROI
How a Premium Bespoke Tailoring Brand Achieved 4.5x ROI Across Digital Channels
We help Indian D2C brands grow with performance marketing, AI automation, and AEO-ready content. Book a free strategy call and we'll show you where the biggest wins are.