Agentic Commerce for Magento 2 - User Guide

Make your Magento 2 / Adobe Commerce store discoverable, readable, and transactable by AI shopping agents. The extension implements four open standards - UCP, Generic MCP, WebMCP, and OpenAI ACP - over a unified catalog service, ensuring every agent sees storefront-correct data (price, currency, tax display, stock, and visibility).

composer require meetanshi/module-agentic-commerce --ignore-platform-reqs
composer require stripe/stripe-php --ignore-platform-reqs

php bin/magento module:status Meetanshi_AgenticCommerce --ignore-platform-reqs

php bin/magento module:enable Meetanshi_AgenticCommerce --clear-static-content
php bin/magento setup:upgrade

php bin/magento setup:static-content:deploy

php bin/magento setup:static-content:deploy -f

php bin/magento cache:flush

composer require meetanshi/module-agentic-commerce

bin/magento module:enable Meetanshi_AgenticCommerce
bin/magento setup:upgrade
bin/magento setup:di:compile
bin/magento cache:flush

Extension Configuration

To configure Magento 2 Agentic Commerce extension, log into the Magento 2 admin panel and navigate to Stores > Configuration > Meetanshi > Agentic Commerce.

General

• Enable Agentic Commerce: Master switch. Set to Yes to enable all agent surfaces globally. When set to No, all agent endpoints return a 404 Not Found regardless of individual surface settings.
• Debug Logging: Set to Yes to write detailed request/response logs to var/log/meetanshi_agentic.log. All secrets and tokens are automatically masked in the log output.

HTTP Surfaces

• Enable UCP Merchant Profile: Set to Yes to serve the UCP merchant profile at /.well-known/ucp. This allows AI agents to discover your store's capabilities and endpoints automatically.
• Enable Generic MCP Endpoint: Set to Yes to serve the Model Context Protocol JSON-RPC endpoint at /agent/mcp. Chat-based assistants (like Claude Desktop or Claude Code) use this to browse your catalog and build carts.
• Enable Public JSON Read API: Set to Yes to serve catalog read endpoints under /agent/*.json. Provides search, product detail, and category listing in a simple JSON format.
• Enable Agent Instructions: Set to Yes to serve /llms.txt and /agents.md discovery documents. These files help AI agents understand your store's category structure, featured products, and policies.
• Enable Agent Discovery in Sitemap and Robots: Set to Yes to inject /sitemap_agentic_discovery.xml into your sitemap and add relevant blocks to /robots.txt, making your store discoverable by AI agent crawlers.
• Enable WebMCP: Set to Yes to inject in-browser MCP tools on storefront pages. Supports both Hyvä 1.3+ (vanilla JS) and Luma (RequireJS) themes with automatic detection.
• Enable Agent Cart: Set to Yes to activate MCP cart tools and the /agent/cart/{maskedId} handoff URL. This allows agents to build carts and hand them off to shoppers for checkout.

Agent Cart

• Continue URL Landing Page: Select where the shopper lands when they open the continue_url from an agent-built cart. Options:
  • Cart - The shopper is taken to the shopping cart page to review items before proceeding.
  • Checkout - The shopper is taken directly to the checkout page to complete their purchase.
• Agent Quote Lifetime (hours): Enter the number of hours an inactive agent-built quote remains valid before being automatically cleaned up. Valid range: 1 to 720 hours. Expired quotes return a 404 when accessed via the continue_url.

Agent Instructions Template

• Policy Pages: Select the CMS pages to expose as variables inside /llms.txt and /agents.md templates. These pages typically include return policies, shipping information, and terms of service that agents can relay to shoppers.
• LLMs Template: Enter the Markdown text template containing Liquid-style variables. This template defines the content of your /llms.txt file and controls what information AI agents see about your store, including category structure, featured products, and policy content.

ACP Checkout Sessions (OpenAI)

• Enable ACP Checkout Sessions: Set to Yes to serve checkout session endpoints at /checkout_sessions*. Required for OpenAI ACP integration.
• API Key (Bearer Token): Enter the authorization secret that the agent platform must send in the Authorization: Bearer header. This value is stored encrypted and never logged in the clear.
• Shared Secret: Enter the shared secret used for HMAC-SHA256 request signature validation. Both your store and the agent platform must use the same secret. Stored encrypted.
• Enable Signature Validation: Set to Yes to require the agent platform to send a valid Signature header on every ACP request. The signature is computed as Base64(HMAC-SHA256(raw request body, shared secret)).
• API Version: Enter the API version validation string. Requests with a mismatched API-Version header are rejected.
• Timestamp Tolerance (seconds): Set the maximum allowed clock drift between the client and server in seconds (valid range: 30 to 3600). Requests with a Timestamp header beyond this tolerance are rejected to block replay attacks.
• Session Retention (days): Enter the number of days to retain completed and cancelled ACP sessions before they are purged by the cleanup cron job.

ACP Product Feed

• Enable ACP Product Feed: Set to Yes to enable the product feed at /agentic/feed. Agent platforms (like OpenAI/ChatGPT) ingest this feed to index your catalog.
• Feed Format: Select the export format for the product feed. Options: csv, tsv, json, or xml.
• Include Categories: Optionally restrict the feed to specific categories. If left empty, all visible products are included.
• Maximum Products: Set a cap on the number of products exported in the feed (valid range: 1 to 100,000).
• Static File Mode: Set to Yes to generate static feed files in var/agentic/ via cron rather than calculating feeds on demand. Recommended for catalogs with more than 10,000 products to avoid performance impact on live requests.
• Feed Regeneration Frequency: Set the cron schedule for static feed file generation. Only applicable when Static File Mode is enabled.

Webhooks

• Enable Webhooks: Set to Yes to fire order.created and order.updated webhook events. Webhooks are triggered only for agent-originated orders (ACP orders and continue_url orders).
• Endpoint URL: Enter the secure HTTPS URL where webhook payloads will be delivered. The configuration rejects http:// URLs - only HTTPS is allowed.
• Signing Secret: Enter the HMAC SHA-256 signing secret used to sign outgoing webhook payloads. The signature is attached as the X-Agentic-Signature header. Stored encrypted.

Delegated Payments (Stripe)

• Enable Stripe Delegated Payment: Set to Yes to allow ACP checkout sessions to charge customers via Stripe's delegated payment flow.
• Stripe Secret Key: Enter your Stripe secret key (sk_live_... for production or sk_test_... for testing). This value is stored encrypted and never exposed in logs.
• Stripe Test Mode: Set to Yes to force all transactions through Stripe's test endpoint. Use this during development and integration testing.

Agent Activity (Telemetry)

• Record Agent Activity: Set to Yes to record agent queries, search terms, cart creations, and order attributions. This data powers the Agent Activity dashboard under Sales.
• Retain Activity For (days): Enter the number of days to retain telemetry records before they are automatically pruned by the daily cleanup cron job.

Product Attributes for Agents

The extension adds three store-view-scoped product attributes under the Agentic Commerce attribute group. These are available when editing any product in the admin catalog (Catalog > Products > Edit).

• Include in Agent Instructions: Set to Yes to include this product in the products.featured array inside /llms.txt and /agents.md templates. A maximum of 50 products can be featured. Use this to highlight key products that agents should proactively recommend.
• Enable in Agent Search: Set to Yes (default) to allow this product to appear in agent search results, category lookups, and direct endpoint views. Set to No to hide the product from all agent surfaces while keeping it visible to regular shoppers.
• Enable Agentic Checkout: Set to Yes (default) to allow agents to add this product to carts and ACP checkout sessions. Set to No to prevent agent-initiated purchases of this product while still allowing it to appear in search results and collections.

Managing Agent Activity

Navigate to Sales > Agentic Commerce > Agent Activity to monitor how AI agents interact with your store.

Agent Activity Dashboard

The Agent Activity dashboard provides a real-time overview of agent interactions with your store, including:

• MCP Endpoint URL: Copy the MCP endpoint URL to configure in your agent platform (e.g., Claude Desktop). The endpoint is your store's base URL + /agent/mcp.
• Funnel Visualization: See the progression from agent queries → product views → cart creations → completed orders.
• Top Queries: View the most frequently searched terms by agents, with unmet-demand highlighting for queries that returned zero results. Use this to identify catalog gaps or search optimization opportunities.
• Attributed Revenue: Track the total revenue generated through agent-originated orders.
• Surface Breakdown: See which agent surfaces (MCP, ACP, JSON API, WebMCP, UCP) are receiving the most traffic.

Test With an Agent

Click the Test with an agent button to run a built-in test that simulates a deterministic agent session:

1. Performs a search_catalog query
2. Fetches a get_product detail
3. Creates a cart with create_cart

The test results are displayed immediately and are excluded from your metrics. Use this to verify your MCP endpoint is working correctly before connecting a real agent.

Managing Webhooks

Navigate to Sales > Agentic Commerce > Agentic Webhooks to view and manage outbound webhook deliveries.

Webhook Queue Grid

The webhook grid displays the following columns:

• ID: The unique webhook queue entry identifier.
• Event Type: The event that triggered the webhook (order.created or order.updated).
• Order ID: The Magento order entity ID (linked to order view).
• Order Increment ID: The order increment identifier displayed in admin grids.
• Store: The store view in which the order was placed.
• Status: The current delivery status:
  • Pending - Awaiting delivery or retry.
  • Sent - Successfully delivered.
  • Failed - All 5 delivery attempts exhausted.
• Attempts: Number of delivery attempts made (capped at 5).
• Last Error: The masked error message from the most recent failed delivery attempt.
• Next Retry At: The scheduled timestamp for the next delivery retry (uses exponential backoff).
• Created At: When the webhook event was first enqueued.