Commerce assistants are moving from simple FAQ search toward structured agent capabilities. On November 11, 2026, Shopify deprecated the Storefront MCP tool search_shop_policies_and_faqs in favor of the UCP MCP ask_business tool. Shopify says the older tool will be maintained until January 15, 2027, but documentation will point developers to the new approach.
For SMBs using AI to answer shipping, returns, services, or store-policy questions, the migration is more than an endpoint change. The response format changes, agent profiles must declare the proper capability, and buyer context can influence the answer. A careful migration should preserve accuracy, identify questions that require a person, and connect the answer layer to the broader customer-response workflow.
What changed in Shopify UCP MCP
The deprecated Storefront MCP tool used the store’s MCP endpoint. Shopify’s replacement uses the UCP MCP endpoint and exposes ask_business, which implements the UCP ask capability. Requests must include an agent profile declaring dev.ucp.common.ask. The question is sent in ask.query, with optional context such as the buyer’s country or preferred language.
The response also changes. Shopify notes that ask_business returns a written answer in structured content rather than raw knowledge-base facts or complete policy text. Existing code that parses the old payload must be updated. Developers should also use tools/list to confirm whether the store exposes the capability for a given agent profile. The official Shopify deprecation notice contains the migration details.
Why the response format matters to a business
A written answer is useful for a customer-facing assistant, but it should not be treated as permission to remove controls. The business still needs to know which policy source is authoritative, when the source was last reviewed, and what should happen when the answer is uncertain or the customer’s case falls outside the standard policy.
For example, “Can I return this?” may depend on product type, purchase date, item condition, market, and whether the product was customized. A fluent general answer can be misleading if the workflow does not collect the necessary context. The response system should distinguish an informational question from a case-specific decision.
Design the migration around Respond → Qualify → Act → Handoff
Respond with the store’s configured information
The first job is to acknowledge the question and provide an answer grounded in the store’s policies and services. Keep the wording direct, indicate material conditions, and avoid inventing exceptions.
Qualify the request when context changes the outcome
Ask only for information needed to determine the next step: order timing, product category, destination country, service type, or preferred language. Avoid collecting sensitive information in an open chat when a secure order workflow is required.
Act through a configured workflow
An action could be presenting the relevant policy link, starting a return request, recording an inquiry, offering a callback, or sending a structured summary to a service queue. Whether an action is available depends on the store’s integrations and business rules.
Hand off exceptions clearly
Complex cases should reach a person with the question, collected context, answer already provided, and reason for escalation. A human should not have to reconstruct the conversation from scratch.
Implementation plan for an SMB
Audit current tool usage
Locate every service that calls the deprecated tool: website chat, shopping assistants, internal support tools, mobile experiences, and test environments. Record endpoint configuration, authentication, parsing logic, timeouts, fallback behavior, and monitoring.
Update capability and request structure
Configure the new endpoint, declare the ask capability in the agent profile, and map the question and optional buyer context correctly. Do not assume a store exposes the tool; check discovery results and design a fallback for unavailable capabilities.
Replace old response parsing
Update code to read the written answer from the new structured response. Log schema errors without storing unnecessary customer data. If another system expects raw facts, decide whether it should instead receive the final answer, a policy URL, or a human-review flag.
Build a representative test set
Create questions covering returns, shipping, pickup, warranties, subscriptions, service areas, languages, and ambiguous cases. Include adversarial or confusing wording. Review whether the answer is accurate, appropriately conditional, and clear about next steps.
Practical SMB example
Imagine a hypothetical home-goods retailer whose website assistant answers delivery and return questions. A shopper asks whether an oversized item can be returned from another country. The new ask tool can provide the relevant store answer, but the business workflow should still capture the destination and product category, show the applicable conditions, and offer a human review if the item falls into an exception.
The retailer tests standard and exception questions, routes uncertain cases to support, and includes a concise conversation summary. It does not promise approval. It creates a controlled path from a general policy question to the right business decision.
How to measure quality
Measure more than containment. Track answer acceptance, repeat-question rate, escalation rate, incorrect-answer corrections, tool availability, response time, and the percentage of handoffs that include the required context. Review a sample of conversations regularly and classify issues by missing policy content, unclear wording, insufficient qualification, integration failure, or business-rule ambiguity.
Also monitor customer outcomes: whether the person reached a next step, whether the case was reopened, and whether a human had to ask for information already provided. A lower escalation rate is not automatically better if customers receive incomplete answers.
Connect store answers to a managed response system
Maya is DIGIMAR’s managed AI Customer Response System for SMBs. Where configured, it can support website chat, inbound phone answering, SMS communication, qualification, appointment or callback workflows, structured summaries, CRM handoff, follow-up, and human escalation. The goal is not a generic chatbot; it is a business-specific Respond → Qualify → Act → Handoff process.
Learn more about the Maya AI Customer Response System, review Maya pricing, or explore custom AI automation and integrations. If your store uses the deprecated Shopify tool, begin with an endpoint and response-schema inventory, then migrate with test questions, escalation rules, and measurable acceptance criteria before the January deadline.