Per-release notes for MCP tools, feed exports, BundleBuilder, post-purchase, substitution engine, and platform integrations. Newest first.
Notify me when changelog updates ships
One short email per shipped feature. Not a marketing drip, just the ship notes.
One email when it ships. No drip campaign.
Platform
OroCommerce integration ships
Composer-installable Symfony bundle for OroCommerce. REST proxies for recommendations and substitutions, admin datagrids for Opportunities, Substitutions, and Jobs, system configuration, and Stripe webhook tier mapping. Hosted-thin-client architecture: the mining engine runs on the MBA hosted backend; the Oro module is a thin wrapper. B2B-first positioning, with RFQ pre-fill via QuoteBundleProposer and reorder predictions for repeat-customer ordering.
Closed the platform gap: Sentry observability, /healthz endpoint, ai_catalog cold-start mining engine, inventory and price enrichment on /recommendations, storefront FBT widget via the BigCommerce Scripts API, and customer-facing Bundle Builder. BigCommerce is now feature-equivalent to Shopify for engine coverage, observability, and storefront integration.
WooCommerce and Magento now support seamless Pro tier activation via Stripe subscription. Subscribe through Stripe-hosted checkout and the plugin activates Pro automatically, no more manual license-key paste from receipts. HMAC-SHA256 signature verification, 5-minute replay tolerance, Pro-wins-on-conflict for multi-item subscriptions. WooCommerce ships with 18 unit tests; Magento ships with 9. Both green.
find_substitutes lands on Shopify, WooCommerce, and Magento
Substitution engine ships as a first-class MCP tool. Given an OOS or unavailable SKU, returns a ranked list of replacements with similarity scores and reason codes (context_similar, category_match, vendor_match). Built for B2B procurement agents resolving unavailable SKUs without escalating to a human.
Given a customer ID, proposes a recurring subscription kit tuned to their predicted reorder cadence. Pairs with predict_reorder. Built for supplement, beauty, and CPG verticals where the LTV comes from converting one-off carts into stacks-plus-subscription.
Exports MBA's mined substitution rules into a Google Merchant Center compatible feed (substitutes column on the product feed). Google's UCP + AP2 protocol expects structured substitutes data; we now ship it.
Captures the post-purchase moment with a one-click upsell offer rendered on the Shopify thank-you page. Recommendations come from the same mining layer as PDP cross-sells, so the data is consistent. Closes the Upsell.com / ReConvert / AfterSell surface gap.
Given the seed SKUs in an RFQ or requisition, returns a fleshed-out bundle proposal with cohesion score plus complementary line items. Hooks into Magento + Adobe Commerce B2B's RequisitionList workflow so buyer-side agents get structured answers instead of escalating to a human salesperson.
Recommendations.csv, substitutes.csv, bundles.csv, opportunities.csv, and rules.csv. Operators can pull the full rules layer into BI tooling (Looker, Tableau, Metabase, dbt). Used by mid-market merchants who want recommendation analytics in their warehouse.
Given a customer ID, returns predicted next-purchase dates per product. Built for reorder-reminder agents and subscription-cadence pre-population. WooCommerce + Magento ports in flight.
Insight Agent tools. analyze_basket returns cohesion analysis for an arbitrary multi-SKU cart. score_cross_sell scores a proposed cross-sell against the current basket. Used by merchandiser-facing dashboards and by agents building bundle proposals.
Given a multi-item cart, returns the products most likely to complete a high-confidence kit. Aggregates per-product recommendations by 'pairs with multiple cart items times average confidence.' First Bundle Agent tool.
First MCP tool. npm @marketbasketanalysis/mcp. Discovery Agent's primary recommendation surface. SSRF-guarded base URL, timeout + retry, bearer auth via merchant-minted API key. Works with Claude Desktop, Claude Code, Cursor, Cline, OpenAI Agent SDK.
LLM-driven catalog miner using a BYO Anthropic or OpenAI key. Solves the cold-start problem for stores with no order history yet. Reasons over catalog metadata to seed plausible accessory + complement pairings. Disabled by default.
Revenue-weighted mining engine. Ranks rules by profit contribution, not raw co-purchase frequency. Critical for verticals where high-margin accessories matter more than low-margin core SKUs (equipment + accessories, beauty finishers, home decor accents).
FP-Growth mining algorithm. Better at long-tail catalog mining than sql_pairs, especially on catalogs with 10K+ SKUs and moderate order volume. Free tier on WooCommerce; included in Pro on all platforms.
First deterministic mining engine. Pure SQL co-purchase rule extraction with support, confidence, and lift columns. Free tier on WooCommerce. The baseline every other engine is benchmarked against.
score_return_risk, propose_visual_bundle, and seasonal-cohort mining are next. We'll add them here as they land. Subscribe above for one short email per release.