Should I use the MCP server or the REST API?
The rough rule: if the consumer is an LLM, use MCP. If the consumer
is code you wrote, use REST.
What’s the difference between an AI agent calling MCP and an AI agent calling the REST API?
Both work; the AI client experience is what differs. MCP gives the agent a tool list at connect time, with structured input/output schemas and human-readable descriptions. The agent picks tools by name and the LLM doesn’t need any “how to call this API” documentation in its prompt — the protocol handles tool discovery, parameter validation, and result framing. REST forces you to put the API surface into the agent’s prompt (or its function-calling schema), which costs tokens and means every release of the agent needs an updated tool definition. For agents that already speak MCP — Claude Code, Cursor, Claude Desktop, VS Code, Windsurf — MCP is strictly less work.What can each surface do?
Both surfaces now cover the full brand-owner product-entry workflow — an
agent holding an MCP connection with a
products:write/dashboard:write
scoped token can create a product and enter every field a human could enter
through the dashboard UI, without switching to REST. See
MCP Tool Reference for the complete write-tool list.
Webhooks and true bulk operations (multi-hundred-row imports) remain
REST-only — those aren’t a good fit for a single LLM tool call.
How do auth and rate limits compare?
Can I use both?
Yes — and many integrations do. A common pattern:- MCP server for the agent’s read path (lookup, search, resolve).
- REST API for the backend that owns mutations (catalog edits, webhook subscriptions, bulk imports) where you don’t want LLM-induced side effects.
Where do I go next?
MCP Install
One-command install for every supported MCP client.
API Reference
Eighteen domain APIs with interactive playgrounds.
Quickstart
First REST API call in under five minutes.
MCP Quickstart
First MCP tool call in under two minutes.