Demand and visibility
LivePulls prompt-level AI engine visibility exports, diffs them against live product and content pages, generates page-level changes, and pushes theme edits through the Shopify CLI.
The super agent for DTC commerce
Shapelab plugs every tool you run your store on into one agent through a single MCP: store, ads, email, analytics, marketplaces, even Canva and Figma for creative. It watches everything, decides what to change, and ships the change. You approve from your phone.
One MCP · every ecommerce tool · approvals from your phone
Meta Ads 3 ad sets fatigued. New variants built in Figma from your brand kit.
ApproveCanva 6 creatives resized for Stories and Reels.
PreviewInventory Reorder 1,200 units by Friday to cover a 4-week lead time.
ReviewEmail Win-back subject lines rewritten. Claim check passed.
ShippedProduct preview · sample data
Super agent
A DTC brand runs on eight dashboards, eight logins and eight sets of exports nobody reconciles. Shapelab puts one agent in front of all of them. It holds the context across every system, so a drop in Amplitude, a segment in Klaviyo and a listing in Merchant Center are one conversation, not three tickets.
Before
8 logins · 8 exports · context lost between every one
After
# one agent, every system as a tool agent: shapelab planner: claude-opus workers: claude-sonnet mcp_servers: - shopify-admin # graphql, read/write - klaviyo # flows, segments, templates - amplitude # events, cohorts - merchant-center # feeds, diagnostics - seller-central # listings, inventory - meta-ads - microsoft-ads - peec-ai # AI engine visibility guardrails: [tone-spec, claim-check]
The system
Each agent runs on a schedule, keeps its skill definition in version control, and writes back to the same systems it reads from. No exports to a spreadsheet, no report waiting for someone to act on it.
Every commercial system is wrapped as an MCP server, so agents query real orders, segments, events, listings and ad spend.
Claude Opus does the planning and the judgement calls. Claude Sonnet handles the high-volume passes: diffs, audits, drafts at site scale.
Theme edits ship through the Shopify CLI, flows through the Klaviyo API, listing files to the marketplaces. Each write is versioned and reviewable.
| Runtime | Claude Agent SDK, cloud-hosted sessions |
|---|---|
| Planning | Claude Opus: decomposition, judgement, approvals |
| Execution | Claude Sonnet: diffs, audits, bulk drafts at site scale |
| Integrations | 8 MCP servers wrapping production APIs (GraphQL and REST) |
| Triggers | cron schedules and platform events (orders, inventory, visibility exports) |
| Skills | versioned in git, changed by pull request, reviewed like code |
| Write path | Shopify CLI theme pushes, Klaviyo API, marketplace listing files |
| Verification | post-write checksum against the live store; refused files surface, not vanish |
| Guardrails | shared tone spec + pre-send claim check (MoCRA, FDA labelling, EU CPNP) |
In production
These run against a real store with real inventory and real regulatory exposure. They are the proof, not the pitch.
Pulls prompt-level AI engine visibility exports, diffs them against live product and content pages, generates page-level changes, and pushes theme edits through the Shopify CLI.
16 flows and 61 always-on emails, specified as structured definitions and built against the Klaviyo API. Customer state is modelled as dynamic segments computed from Shopify events, not stored properties.
Social reply triage, PR outreach and publishing across four platforms. Every message passes a shared tone specification and a pre-send claim check, because cosmetic claims are regulated.
Marketplace listing file generation, INCI and compliance data, and image filename and alt text audits across the whole site.
Results
Runs from your phone
The agents run in the cloud, not on a laptop that has to stay open. Scheduled jobs fire whether you are at a desk or not. When one needs a decision, it asks on your phone: approve a rewrite, preview a theme change, redirect a run, or ask it anything about the business in plain language.
Example conversation
Next
The next agents make the decisions that cost the most when they are wrong: spend, demand and stock.
The agent reasons about a budget ceiling across Meta and Microsoft Ads instead of obeying a static cap. It moves money toward what is working and explains why.
Behaviour and demand forecasts from first-party event data in Amplitude and Shopify, not third-party panels.
Inventory reorder quantities sized against three to four week manufacturing lead times, so you reorder before the stockout, not after it.
One dashboard, read and write. It is the agent's interface, not a report. You see what it plans to do, approve or adjust it, and the change goes out from the same screen.
Join the waitlistTenant zero
koppa beauty
koppa has inventory, regulatory exposure and customers in six markets. It is exactly the problem we are solving: genuine operational complexity and no headcount. Every agent runs on koppa before it runs for anyone else.
Who builds it
Built by an operator who has shipped software at scale.
Our founder is also Head of Product at My Study Life, a cross-platform student planner with 24 million lifetime downloads across 197 countries.
Early access
We are opening Shapelab to a small number of operators with the same problem we had. Tell us about your store and we will be in touch.
Or email hello@shapelab.me
FAQ
An agent. The dashboard on the roadmap is how you read and steer it, but the work happens on a schedule whether you open it or not.
Shopify Admin GraphQL, Klaviyo, Amplitude, Google Merchant Center, Amazon Seller Central, Meta Ads, Microsoft Ads and Peec AI today. Each is wrapped as an MCP server, so new systems are added the same way.
Outbound messages pass a shared tone specification and a pre-send claim check. You decide which pipelines write automatically and which wait for approval.
Cosmetic claims are regulated: MoCRA and FDA labelling rules in the US, CPNP in the EU. One wrong word in a reply or an email is a compliance problem, so every claim is checked before anything goes out.
No. The agents run in the cloud on their own schedules. You approve, preview and redirect them from your phone.
DTC operators with real operational complexity, such as inventory, several markets and regulated products, and not enough people to run a growth team.
The production pipelines run on koppa beauty now. We are onboarding a small group of operators ahead of the spend, prediction and reorder agents. Request access above.