MCP Servers for 22 Ecommerce Systems

Model Context Protocol (MCP) is a standard way for an AI agent to work with other software: an MCP server turns a system's API into a set of structured tools the agent can read from and act through.

Between 17 and 19 July 2025, Vortex IQ released MCP servers for 22 systems that sit around an ecommerce store, from analytics and email to ERPs, helpdesks and file transfer. A store's data is spread across many tools, each with its own API and login. With an MCP server in front of each one, an agent could ask Google Analytics a question, check stock in OrderWise and draft a Klaviyo email through one standard interface, without a custom integration for every pair of tools.

Each server was announced in its own post at the time. This page brings those 22 announcements together in one place, grouped by the job each system does. If you followed a link to one of the original posts, you have landed on the section for that system. Most of the announcements described the same base: authenticated API access, role-based access control, audit logs of agent actions, and a choice of hosted or self-managed deployment.

Analytics and reporting

Three servers dealt with numbers a store already collects but often struggles to act on quickly.

Google Analytics

Google Analytics records how people reach a store and what they do once they arrive. Its server, one of three released on 17 July 2025, put the Google Analytics API behind an interface a language model could query in plain English.

A marketer could ask, "What's our bounce rate for mobile users in Germany last week?" and get the answer without opening a report. The agent could also flag a jump in checkout drop-off at the shipping details step. One worked example set an alert for bounce rate on paid mobile traffic going over 65% in a week. The agent compared the live figure with a baseline, posted the alert in Slack and recommended which campaign to pause.

The announcement named marketers, merchandisers, UX teams and developers tracking Core Web Vitals as users, along with executives who wanted a weekly digest.

BI platforms

Plenty of stores report through Looker, Power BI, Tableau or a custom BI setup, then wait hours or days before anyone acts on what the dashboard shows. The BI server turned reports, KPIs and saved queries into tools an agent could read and monitor.

Questions came in plain language: "Why did our AOV drop yesterday?" or "Send a summary of low-performing SKUs this week." A third listed every product with more than 500 sessions and under 1% conversion that week. The agent could also set sales figures against stock. High stock and slow sales pointed to a promotion. Low stock and rising sales pointed to a reorder, or a pause on the ads driving demand.

Leadership could receive daily, weekly or monthly summaries of revenue, best and worst SKUs and campaign ROAS by email or Slack, while BI analysts and ecommerce managers worked from the alerts.

Magento Analytics

Merchants on Adobe Commerce (Magento) get reporting across sales, customers, marketing and operations, but acting on it has usually meant manual analysis. Released on 18 July 2025, this server worked with Adobe Commerce Open Source and Commerce Cloud, and let an agent query that data in natural language.

Typical requests from the launch:

"Alert me if revenue from mobile users drops more than 15% week-over-week."
A Monday report of the top 10 products by conversion rate, highlighting any that fell more than 20% since the previous week.
A weekly summary of sales by category, top-selling SKUs, refunds and cancellations, and conversion rate by device.

When a product's sales dipped, the agent could check whether stock was the cause. A low-stock item could come off the homepage and out of paid campaigns, while a well-stocked, slow seller could get a clearance offer or bundle. Data analysts got automated reporting; ecommerce managers got alerts and forecasts.

Ads, email marketing and pricing

These servers sat closest to revenue: paid social, email and SMS, and the price on the product page.

Facebook (Meta) Ads

The Facebook (Meta) Ads server came out on 17 July 2025. It converted the Meta Ads API into tools a model could use across ad accounts, campaigns, creatives, audiences and budgets.

Budget rules were the simplest use. A media buyer could set an instruction such as "Pause any ad with ROAS under 1.5 for more than 24 hours." A campaign could be described in a sentence, for example a traffic campaign promoting a summer collection to women aged 25 to 45 in the UK, and the agent would build the targeting, assign the budget and pull product images from the store or CMS. Retargeting ad sets could target shoppers who abandoned a cart or viewed a product several times, and lookalike audiences could be refreshed from purchase data with existing customers excluded.

Support for multiple accounts and brands made it relevant to agencies running campaigns for several clients, as well as to in-house performance teams.

Klaviyo

Many ecommerce brands run their email and SMS through Klaviyo. On 18 July 2025 we released a server that gave agents its engagement data (opens, clicks and conversions) along with its segments, flows and campaigns.

Abandoned carts were the obvious starting point. One example sent a recovery message when a cart was worth over £100 and the shopper had not come back within two hours, once the agent had confirmed the products were still in stock. Other uses went further along the customer lifecycle:

building an "at risk" segment from purchase frequency and email engagement, then starting a win-back journey
following an order with a thank-you, a review request through Feefo or Trustpilot, and a replenishment offer 30 days later
sending back-in-stock or price-drop alerts to people who engaged with a product but did not buy

For CRM and email marketers, segments updated as customer behaviour changed. Retention teams got churn signals they could act on.

Apollo Pricing

What should a product cost today? Apollo Pricing answers that from demand, stock levels, competitor prices, seasonality and customer behaviour. Its server exposed pricing recommendations, elasticity data, rules and forecasts to agents, so prices could respond to signals from GA4, Shopify, Magento, Klaviyo and OrderWise.

Most of the examples came with limits attached. An agent could raise a price within a pre-approved range when product page views jumped and stock was limited. It could match a competitor's price only where the margin stayed above target, and warn merchandising when undercutting would hurt profit. For slow stock, it combined Apollo's suggestions with OrderWise stock levels: one example applied a 15% discount to items with more than 90 days of stock cover, then passed them to Klaviyo for a promotional email.

Pricing teams set the rules, and finance teams got margin monitoring. The post also listed audit logs and rollback.

Customer service, CRM and reviews

Four servers connected agents to the places where customers talk back: two helpdesks, a CRM and a review platform.

Freshdesk

Answering a Freshdesk ticket usually means looking something up elsewhere: an order, a shipment, a product. The Freshdesk server, released on 18 July 2025, let an agent sort incoming tickets by intent, urgency and sentiment, then pull the data needed to answer them from Shopify, Magento, Klaviyo or OrderWise.

Take "Where is my order?" The agent could look up the shipment in OrderWise or ShipStation, reply with the tracking link and delivery estimate, and escalate only when the parcel was late or returned. Return requests could be checked against the return window before instructions went out. Another example escalated low-CSAT tickets from VIP customers to senior support within 10 minutes.

Product teams benefited too. When customers kept reporting the same fault on one SKU, such as "size too small", the agent flagged it to merchandising and could suggest advisory text for the product page.

Zendesk

Zendesk's server translated ticket data, user history, macros, triggers and workflows into a form an agent could read and act on.

Routing came first. An agent could read each new ticket, work out its topic, urgency and sentiment, and send it to the right queue. A launch example tagged any ticket mentioning "damaged item" and assigned it to the returns team when the order was worth more than £100. Order and shipping questions could be answered with live status from OrderWise, Shopify or Magento.

The announcement also described an assist mode for human support staff. During a conversation, the agent suggested macros, knowledge base articles and next steps, and surfaced related tickets from the same customer. Replies get faster while a person stays in the conversation. The agent could also spot a sudden rise in tickets about "late delivery" or "checkout error" and tell CX, logistics or developers.

Salesforce

Leads, opportunities, cases, contacts and accounts are the core objects in Salesforce. The 19 July 2025 server turned each one into an endpoint an agent could read, create and update, and connected them with Shopify, Magento, Klaviyo, OrderWise, GA4 and Freshdesk.

Store data made room for sales and service tasks:

Abandoned carts worth over £200 became Salesforce leads, with a follow-up task for the SDR team.
Opportunities stalled for more than seven days were flagged for review.
A complaint arriving through Freshdesk or Trustpilot was matched to the existing contact and logged as a case.
Contacts with lifetime value over £500 and repeat purchases in Q2 moved into a "Premium Loyalty" segment in Klaviyo.

Because the agent could check OrderWise stock, the cross-sells it suggested in opportunity notes were limited to products actually available. Sales, support and marketing teams each had a use for it.

Trustpilot

Reviews carry detail that rarely reaches the product page. Trustpilot collects and displays verified customer reviews, and its server let agents monitor that stream for sentiment, keywords, ratings and repeated complaints.

One example watched for sizing comments: if more than three reviews mentioned a sizing issue within ten days, add a "check size guide" alert to the product page. Another moved every 1 to 2 star reviewer onto a "CX Risk" list in Klaviyo and told the support team. If the same fault came up three or more times in 48 hours, the agent could alert fulfilment, the product team and the helpdesk, and pause marketing for that product.

Good news travelled too. Products rated above 4.5 stars with more than 25 reviews could get a "Customer Favourite" badge and their best quotes on the page. The agent could also draft brand-aligned replies to new reviews and pass complex cases to a CX manager.

Search, product data and digital presence

These four servers dealt with how products are found and described.

Klevu

A search that returns nothing is a dead end for the shopper. Klevu, which handles on-site search and product discovery for retailers, got its MCP server on 18 July 2025. Agents could query search terms, trends and performance, then refine merchandising rules, boosting and synonyms.

Zero-result searches were the first target. A rule from the launch: if more than 50 users search for a term with zero results in 24 hours, add a synonym or suggest similar products. A fuller example imagined 100 shoppers searching for a new line called "Bamboo Activewear" and finding nothing. The agent would create the synonym, alert merchandising and update homepage banners.

Stock mattered as well. If the top result was out of stock, the agent could reorder results so in-stock items came first. Merchandisers could change boosting and synonyms by describing what they wanted.

Yext

Store hours, FAQs, location pages and local listings need to say the same thing everywhere a customer looks. Yext manages that structured content across a brand's own site, its store locator, and platforms such as Google, Apple and Amazon Alexa. The server released on 19 July 2025 covered Yext's Knowledge Graph, Search, Pages, Reviews and Listings.

Search behaviour came first. When over 100 visitors searched "track my order" and then left, the agent could create an FAQ entry linking to the order status tool, or draft missing FAQ entries for review.

Store data was the other half. One instruction updated opening hours for a bank holiday across every location and pushed the change to Google Maps, Apple Maps and voice assistants. Low ratings at a particular location could be routed to that store's manager. Digital experience, SEO and store operations teams all had a stake.

Chronos (Magento Product IDs)

Before an agent can change a product, it has to know exactly which product it means. Chronos, Magento Product IDs, is a lightweight API that exposes product identifiers and metadata in a Magento store. Its server let an agent resolve a SKU or product name to the right Magento product ID without fuzzy matching on names.

The launch example read: "Update the hero image for 'Blue Trail Running Shoes' and confirm it's applied to the correct product ID." The agent looks up the ID first, then changes that product only. The same lookup helped elsewhere. GA4 events carry the SKU but not the Magento ID, so Chronos let behaviour data be joined to the product record. Returns raised in Freshdesk could be matched to a product ID from order history, and Shopify entries or Google Shopping feed data mapped to Magento's structure.

Developers needed this so agents targeted the right product. Merchandisers could use it to automate product page updates.

Nutritics

Food and wellness retailers carry product data most stores never deal with: allergens, macros, ingredients and dietary suitability. Nutritics is a platform for food labelling, menu analysis, nutrition tracking and dietary compliance. Its server arrived on 19 July 2025, aimed at grocery marketplaces, supplement brands, meal-kit providers and nutrition-focused retailers.

With that data available to an agent, a request such as "Show all snacks with less than 5g sugar per serving and over 7g protein" could be answered from the product data itself. Customers with a nut allergy set in their profile could have nut-containing products removed from search and recommendations. Bundles could be built to a target, for example five products delivering around 1,500 kcal a day with balanced protein and fibre.

Compliance teams had their own use. The agent could fill in nutritional labels, traffic light ratings and dietary badges such as "Suitable for Coeliacs", with UK HFSS rules named as one of the labelling laws to meet.

Inventory, orders, ERP and B2B procurement

This is the largest group. Five servers reached the systems that hold stock, orders, purchasing and finance, and a sixth served B2B buyers who order through procurement software.

OrderWise

OrderWise is a UK ERP and warehouse management system covering inventory, order fulfilment, accounts and purchasing. Its server let agents monitor stock, fulfilment times, returns and purchase orders, and link them with Shopify, Adobe Commerce, BigCommerce, GA4, Feefo and Meta Ads.

Three of the launch instructions show the range:

If stock falls below 20 units and product sessions rise 40% in a week, raise a reorder and flag the product to come off the homepage.
If a product has more than 500 units and fewer than five sales a week, apply a 15% discount and add a homepage banner.
If dispatch times slip, tell CX, update the delivery promise on the site and send a delay email through Dotdigital or Klaviyo.

Returns gave another signal. The agent could read return reasons in OrderWise, compare them with Feefo reviews, and alert product, CX and logistics teams when "damaged" or "not as described" kept appearing. Suppliers could be ranked on delivery consistency, defect rate and unit cost.

Microsoft Dynamics NAV (Navision)

Navision remains a common ERP among mid-sized businesses and is now part of Microsoft Dynamics 365 Business Central. The 18 July 2025 server reached it through OData or SOAP APIs and worked with NAV 2013 onwards, Business Central and hybrid setups.

Finance work sat at the centre of this one. An agent could query accounts receivable, find overdue invoices and customers with high aged balances, and respond with reminder emails, a credit hold flag and an alert to the account manager. It could also report daily or weekly on sales by channel, gross margin by category, aged receivables and cost of goods sold, flagging unusual spikes or margin drops.

Operations had examples too. One reordered any SKU with fewer than 20 units in stock when its seven-day traffic was up 30%. Another checked high-value orders against credit limit, pricing and stock, and held any order with a discrepancy for review. Finance teams and inventory managers were the first audience.

Reflex Prod

Getting ERP data into daily decisions often means spreadsheet exports and manual reports. The announcement for Reflex Prod, an ERP and warehouse management platform used by retailers and manufacturers in Europe, said as much.

Released on 19 July 2025, the server let an agent watch sales, stock and fulfilment in Reflex Prod alongside Google Analytics, Magento and Shopify. One rule: reorder any item with a seven-day sales spike above 20% and stock under 25 units. Where Google Analytics and Magento showed a regional surge, the agent could check store-level stock and suggest a transfer, with logistics notified to approve or schedule it. Courier delays on particular SKUs could prompt an email or SMS to affected customers and an alert to CX with the new delivery date.

Multi-language and multi-store setups were supported. Inventory managers and ecommerce operations teams were the core users, with BI teams forecasting demand for peaks such as Black Friday.

Futura Inventory Update

Retailers with stock spread across several warehouses, stores and sales channels use Futura Inventory Update to track product movement and availability. Its server gave agents stock levels, turnover rates and availability to read next to traffic and campaign data.

Campaigns felt it first. One instruction paused any campaign featuring a product with fewer than 10 units available. When GA4 showed a jump in views or add-to-carts, the agent could check Futura stock against replenishment timelines and, if a sell-out looked likely, raise a reorder request, notify the buying team and show a "low stock" banner.

Slow stock went the other way. Products with more than 500 units and a sales rate under 3% over 30 days could get clearance pricing and a place in the next Emarsys or Dotdigital email. Restocked items could trigger back-in-stock messages by email or SMS. Buying teams and inventory managers sat at the centre of it.

Inventory Planner

Forecasting is the focus of Inventory Planner, a replenishment platform that helps merchants predict demand, reduce stockouts and improve cash flow. The 18 July 2025 server made those forecasts available to agents alongside Shopify, Magento, OrderWise, Klaviyo, GA4 and ERP data.

The launch examples built on the forecast:

reorder any product forecast to sell out within 15 days where supplier lead time is longer than seven days, then notify procurement
pause a Klaviyo or Meta promotion if the featured product has fewer than 10 units and no incoming purchase order, and suggest an alternative
score suppliers on delivery times and fill rates
when a purchase order is delayed, alert the CX team through Freshdesk and update the delivery estimate on the product page

Inventory planners and procurement teams were the main users, and marketing got a way to keep campaigns in line with stock.

PunchOut Gateway

B2B buyers at large organisations often order through procurement systems such as SAP Ariba, Coupa, Oracle or Jaggaer. PunchOut Gateway connects an ecommerce store to those systems. As the original post noted, PunchOut catalogues are usually static, with little view of what buyers do in a session.

The 19 July 2025 server let agents read PunchOut session data and act on it. When a buyer added items to a cart but did not complete the requisition, the agent could send a follow-up with a cart summary, alert the assigned account manager and optionally prepare a quote. When a buyer went over a budget or approval threshold, it could start a custom quote and route the request to the right person. Catalogue views could be tailored to each account's purchase history, approval levels and contract pricing.

For a B2B seller's sales team, that meant buyer intent alerts and quotes ready for follow-up.

File transfer and email delivery

The last two servers dealt with plumbing: the files that move between systems and the emails a store sends.

SFTP

Order exports, inventory updates, supplier catalogues and compliance reports still move between businesses as files on SFTP servers. The SFTP server, released on 19 July 2025, let an agent read, write and move files in secure directories, act when a file arrived, changed or fell due, and parse CSV, XML, JSON and XLSX content.

A daily inventory CSV from an ERP could be parsed and used to update stock in Magento, Shopify or BigCommerce. Orders could be validated (paid, with a verified address) and exported every hour to a 3PL's folder. Supplier catalogue files could be checked for required fields and mapped to Magento or Shopify format. If an expected file, such as a product feed due by 7am, was missing or malformed, the agent could raise an alert in Slack, by email or as a Freshdesk ticket.

SSH key and password authentication and PGP encryption were supported. Operations teams, developers and finance were the main users.

SMTP

Underneath much of the transactional and campaign email a store sends sits SMTP, the Simple Mail Transfer Protocol. This server worked with any SMTP-compatible service, including SendGrid, Mailgun and Amazon SES, and with stores built on Laravel, Magento, Shopify or WooCommerce.

An agent could send an order confirmation as soon as an order completed, then delivery updates as the parcel was dispatched, out for delivery and delivered. It could follow up abandoned carts worth more than £50 after two hours, once it had confirmed the items were still in stock. It also read delivery logs. If bounces or spam flags formed a pattern, the agent could flag the affected recipients, pause the campaign and retry with changed content or another provider.

After delivery, a review request could go out through Feefo or Trustpilot. CX teams, developers who wanted fewer custom scripts, and marketers who wanted behaviour-driven email without depending on a marketing platform were all named as users.

Where these connections fit today

The 22 posts gathered here date from July 2025, and several of their examples describe an agent acting on its own: pausing a campaign, changing a price, replying to a customer. Read them as a record of what each server was built to reach.

Today, Vortex IQ works across 200+ connectors. Every change an agent prepares goes to a person for approval first, whether it is a price, a paused campaign or a reply to a customer. It proposes. Your team approves. Nothing goes live without your say-so.

The 22 MCP Servers We Released for Ecommerce in July 2025