Tested playbooks for product feeds, Google Shopping, GA4, catalog translation and checkout. Hand one to a hosted agent: it does the work and reports back where your team already talks.
Drag the slider. On the left, titles from a typical export; on the right, after the skill: the attributes people search for first, no promo text and nothing invented.
BeforeAfter
8841SALE!!! Mens shoes Speedcross 6 GTX best price
8842Merino socks 3pcs FREE SHIPPING
8843Backpack 28L blue hiking backpack for trips backpack
8844Thermos 0.75 TOP
8841Salomon Men's Trail Running Shoes Speedcross 6 Gore-Tex Black 43
8842Merino Hiking Socks Crew 3-Pack Grey L
8843Osprey Hiking Backpack Talon 28 L Blue
8844Stanley Classic Stainless Steel Thermos 0.75 l Green
Apparel & footwear: Brand · Gender · Type · Material · Colour · Size
The first 70 characters carry what shows in the results
No "sale", "free" or ALL CAPS — those get disapproved
A missing attribute stays missing and the row is flagged
Warns when the product page H1 disagrees with the title
Write your brand voice, shipping and margins into the Brain. The agent reads them on every task, so you stop repeating yourself.
# memory/brand-voice.mdvoice: informal, factual, no superlatives
shipping: free from 1,500 CZK
margin: minimum 28 %
Setup
From sign-up to a working agent in four steps.
No server and no code on your side. You click, Eco does the rest.
1
Sign up on ecomdly.com
With your Google or GitHub account. No card.
Continue with Google
2
Create an agent
Give it a name and pick the model: try ours free in the demo, or bring your own API key.
eco-merinoLauni AI · demoOwn API keyStart agent
3
Connect Slack or Discord
Pick the channel where Eco reports. You can change it any time.
SlackDiscord
✓
Done. Pick the skills
Choose them yourself, or let Eco take one from the catalog when a task matches it.
weekly-store-kpi-reportproduct-feed-optimizer“Why are 40 products disapproved?”merchant-center-disapproval-fixer
1 month free, everything included · 394 of 400 free trial agents left
Break-even ROAS calculator
How much must ads bring in for you not to lose money?
From the contribution margin after returns and fees. The break-even-roas-calculator skill does the full itemised calculation.
Break-even ROAS3,58
Max PNO28 %
Target ROAS with profit5,01
Why trust it
No skill reaches the catalog without review.
The agent works from exports and with read-only access. Changes to your feed or prices are only ever proposed, and you approve them. The catalog is mirrored hourly to a public GitHub repo.
Markdown and license checkpassed
Secrets and hidden-instruction scanpassed
Safety scan: deceptive practices, data deletionpassed
Human review and versioningapproved
Pricing
Free to start. Pricing is on its way.
We are preparing the paid plan. Until then the hosted agent is free for a month with everything included, for the first 400 agents.
Hosted agent1 month freethen Pro — pricing coming soon
Eco works in your Slack or Discord
On a schedule or when you ask
Our model to try in the demo, or your own API key
394 of 400 free trial agents left
Questions
What store owners ask
Does the agent get access to my store?
No. Skills work from the exports you give them (feed, GA4, orders). A hosted agent can get read-only access to specific data, revocable at any time. It only ever proposes changes.
What does it cost?
Browsing and installing skills is free. The hosted agent is free for a month with everything included, for the first 400 agents. Pricing for the paid plan is being prepared.
What is a skill?
A skill is a Markdown file that tells an AI agent how to do one e-commerce job: when to use it, what data it needs, the rules to follow and the format of the result. Every skill in the catalog passed automated checks and a human review.
What is the hosted agent?
Eco, an agent we run for you. You pick its skills and it works in your Slack or Discord, on a schedule or when you ask, and brings results for you to approve. It runs on our model, which you can try free in the demo, or on your own API key.
Which platforms does it work with?
Any that exports CSV or XML: Shoptet, Shopify, WooCommerce, Upgates or a custom build. Skills know the feed formats of Merchant Center, Heureka and Zboží.cz as they are.
Do I need to code?
No. In Claude one command is enough, or you start a hosted agent from the browser. A skill is readable text, so you see exactly what the agent does.
Where is my data?
ecomdly and your Brain run on our own server in the EU, and Brain files are encrypted at rest. Sign-in goes through Google or GitHub, so we hold no passwords.
What if the agent gets something wrong?
Changes to your feed, prices or ads are only proposed and nothing is applied until you approve it. Skills are written to flag missing data instead of inventing it.
Do I need a Claude account?
Not for the hosted agent. To use skills in your own agent you need that agent, for example Claude Code or any client that speaks MCP.
Give an agent its first task today.
Pick a skill, upload an export and have a result to review in minutes. Free, no card.
The catalog
Every skill, reviewed and free to use.
E-commerce skills for AI agents — playbooks for product feeds, Google Shopping, GA4, catalog content and checkout. Browse, like, and add them to Claude in one command. Everything published is mirrored to GitHub.
claude mcp add --transport http ecomdly https://ecomdly.com/mcp
Reads out an e-commerce A/B test in the right order: SRM chi-square, pre-registered primary metric, sample size and peeking, effect with confidence interval, guardrails and novelty, then ship, don't ship or inconclusive.
Plans custom_label_0-4 from your own margins, VAT-correct break-even ROAS and sample-safe performance tiers, and outputs a mapping file for a supplemental feed plus the Shopping or PMax structure each label drives. For PPC specialists.
Reviews Performance Max asset groups for Merchant Center retailers: product ownership and overlaps, unserved products, asset gaps against Google's specs, message match, signals, brand exclusions and Final URL expansion, with proposals for a human to approve.
Audits a product page's JSON-LD against Google's merchant listing and product snippet rules and against the visible price, stock, ratings and written shipping/return policy, then lists errors, gaps and confirmed fixes.
Writes a one-page Monday KPI report for store owners: backend revenue, orders, AOV, conversion, MER and refunds versus last week and last year, with noise tests and at most two decisions.
Segments lapsed customers by their own purchase cycle and net-of-returns RFM, plans a sequence and honest offer per segment with a holdout and break-even check, and sunsets unresponsive contacts.
Builds and checks the 301 map for a store moving platform, URL structure or domain: multi-source URL inventory, ID/SKU/EAN matching before fuzzy, discontinued and filter URLs, chains, canonicals, hreflang and post-launch checks.
Lets an agent read and change a Shopify store through the GraphQL Admin API safely: pinned version, least-privilege scopes, cost-aware paging and bulk exports, userErrors checks, owner-approved diffs and a rollback log.
Lets an agent read and write iDoklad through API v3 for an online store: unpaid and overdue invoices, payments backed by bank records, Default-template drafts the owner approves, credit notes and accountant exports.
Lets an agent work an Adyen merchant account safely: find payments, explain refusals, prepare owner-approved refunds, captures and cancels, triage disputes by deadline, and reconcile settlement reports to orders.
You own one deliverable: a machinereadable translation glossary extracted from the live catalogue, splitting every recurring string into terms that must be translated consistently, brand and model nam...
Sets or reviews a free-shipping threshold and shipping prices from your own basket distribution, carrier cost per parcel, contribution margin and bunching below the line, with profit per order and a test readout, no invented lift.
Lets an agent work e-shop Jira tickets like a good colleague: readiness check, one numbered clarifying comment with defaults and an @mention, ADF comments via REST v3, real transitions only, internal-first replies in JSM.
Lets an agent work a store's Stripe account safely via the Stripe plugin, MCP or REST: find payments, explain declines, triage disputes by due date, reconcile payouts and fees, and prepare owner-approved refunds.
Calculates per-SKU, per-marketplace contribution on Amazon EU, Allegro and Kaufland after commission, fulfilment, storage, ads, returns and FX, on the right VAT basis, and decides keep, reprice, change fulfilment or delist.
Turns Amazon Account Health, Allegro sales-quality and Kaufland performance exports into a risk-ranked fix plan: metric, window, slack in orders, roll-off date, root cause, fix and owner per issue.
Reallocates a fixed monthly budget across Shopping and Performance Max campaigns by margin-based break-even, lost IS to budget vs rank and marginal return on matured data, as stepwise moves for the owner to approve.
You own the fieldlevel teardown of a live checkout: every input the shopper must touch between "proceed to checkout" and "order placed", scored for cost, ranked for removal, merge or autofill, and han...
Decides before launch whether a sale, coupon or bundle pays: break-even uplift from contribution margin, pull-forward, cannibalisation, coupon leakage, ad-target impact, EU 30-day prior price check and a baseline readout.
Triages consumer claims for faulty goods under the EU legal guarantee (2019/771) with country-correct deadlines and burden of proof, then drafts the acknowledgement, evidence request and decision letter for the owner to approve.
Designs a welcome flow from signup to first purchase with provable EU consent, double opt-in, exit rules, a margin-tested discount decision, Gmail/Yahoo sender checks and a holdout. Drafts only, never sends.
Audits on-site product search from GA4 view_search_results data or vendor logs plus a hands-on test set (zero results, no-click queries, diacritics, SKU/EAN, filters, mobile) and ranks fixes by lost revenue.
You own the weekly rhythm artifact: a 9slide spine (hard cap 14 including appendix) that tells the store's owner, head of growth and performance buyer what moved last week, why, and what gets decided...
Triages new reviews from Google Business Profile, Heureka, Trustpilot and the store's site by severity and drafts one-standard public replies plus internal follow-ups, policy-checked against platform rules and EU review law.
No passwords. We only read your name, e-mail and avatar.
Two ways, one account
In your own AI
Free
Connect ecomdly over MCP to Claude, Cursor or another AI. It finds and installs skills from the catalog by itself.
ecomdly.com/mcp
An agent of your own
Pro · 30 days free
Works on its own on a server: watches Merchant Center, sends GA4 reports to Slack, files invoices from Gmail.
You don’t need an agent — start one any time later. The catalog works over MCP even without an account; the account adds a private Brain and publishing skills.
Cookies for analytics?
We use Google Analytics to see which skills people find useful. It sets cookies only if you agree; the site works the same either way. What we store.