All posts
ProductJune 5, 2026·8 min read

Agent-Native Commerce: Introducing the PlatformDTC Agent Gateway

By PlatformDTC Team


For two years “AI in commerce” mostly meant a chat box that answers questions. Helpful — but it doesn’t run the store. Today we’re shipping the piece that does: the PlatformDTC Agent Gateway — one secure, programmable API that lets autonomous agents take real actions across your entire store.

Chatbots talk. Agents act.

An agent should be able to build a storefront from a brief, source and import products, publish to a live domain, connect ad accounts, launch a campaign, and watch ROAS — then decide what to do next. Doing that safely against real money and real customers needs more than a model with a prompt. It needs an enterprise control plane.

What the Agent Gateway is

A versioned API at /api/v1/agent/v136 endpoints spanning store building, catalog, ads, marketing, analytics, and orders — mirrored 1:1 by 23 dtc_* tools over the Model Context Protocol, so an OpenClaw agent registers a single server and gets the whole platform.

Least privilege

Per-agent keys carry only the scopes you grant; spend actions are gated.

Safe retries

Idempotency keys are mandatory — duplicate calls never double-charge.

Async by design

Long operations return a job; one surface to poll for every workflow.

Full audit

Every call recorded: principal, tool, scopes, status, latency.

Signed webhooks

HMAC-signed event callbacks so agents react instead of polling.

Human approval

Optional gate on spend — agents propose, humans confirm.

One sequence: brief → store → ads → ROAS

Here’s the headline flow an agent runs. Every long step returns a job; the agent waits on it, then moves on. Spend steps can require a human tap.

dtc_create_store { name:"Aurora Skincare", brief:"minimalist sage skincare, 6 products" }
  → job → dtc_job_wait → { project_id }
dtc_bulk_import_products { project_id, products:[…] } → job → wait
dtc_publish_project { project_id } → job → wait → { production_url }   # store:publish
dtc_connect_ad_platform { platform:"meta" } → { auth_url }            # human authorizes
dtc_create_campaign {…} → dtc_send_campaign {…}                       # marketing:send (may need approval)
dtc_get_dashboard_analytics { range:"7D" } → { revenue, orders, ad_spend, roas }

The exact same calls work over plain HTTP with an API key — anything an agent can do, you can reproduce with curl:

curl -X POST https://api.platformdtc.com/api/v1/agent/v1/projects \
  -H "X-Agent-Key: sq_agt_xxx" -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{"name":"Aurora Skincare","brief":"minimalist sage skincare, 6 products"}'

Safe enough to let it run

Autonomy is only useful if it’s contained. Each agent gets its own sq_agt_* key with least-privilege scopes, an independent rate limit, and instant revocation. Mutations are idempotent. Spend and customer-contact actions (publish, ads, send) can be put behind a human-approval gate, so an agent can propose “spend $5k/day on Meta” without it auto-executing. Every call lands in an audit log.

Built for OpenClaw — open to everyone

The Gateway ships with a PlatformDTC MCP server so OpenClaw agents are productive immediately, and a reference skill that encodes the playbook above. But it’s just HTTP underneath — any agent framework, any language.

Start building

Read the docs, mint a key, ship your first agent.