Your catalog, one npx away from Claude. And a REST API your developers will not hate.
Apimio ships a public REST API over your product catalog, HMAC-signed outbound webhooks, an OpenAPI 3.1 reference with code samples in five languages, and a published MCP server that connects any MCP client to a merchant catalog with 47 tools — running against production, not a sandbox. The same feed engine that serves Google and Meta also produces an agent-ready feed shaped for AI shopping assistants. Agencies automate client catalog work against it. AI tools read product data through it. It is the same catalog your stores and feeds already run on.
The parts a developer actually checks for
The things that decide whether an API is pleasant to build against, rather than the things that look good on a feature list.
REST API v1
Products, variants, media, categories, brands, vendors, attributes, inventory across all eight states, multistore overrides and locales — read and write, with stable contracts that do not leak internal IDs.
- Cursor pagination, not offset
- Idempotency keys on writes
- Bulk endpoints up to 500 products per call with job-status polling
- Filtering across every attribute type
Scoped API keys
Keys are bound to an organization and managed under role-based permissions, with per-plan monthly quotas and rate limits, and admin-grantable access for partners working on a client account.
- Org-bound, RBAC-controlled key management
- Quotas: 250k calls/mo Advanced PIM, 1M Enterprise PIM
- Rate limits scaled per plan
- Partner access grantable by an admin
Signed outbound webhooks
Product and variant events delivered with HMAC-SHA256 signatures so you can verify origin, with automatic retries and backoff, an inspectable delivery log, manual redelivery, and test pings while you build.
- HMAC-SHA256 signed payloads
- Automatic retry with backoff
- Delivery log with manual redelivery
- Test pings during development
A reference somebody proofread
A branded API reference generated from a hand-authored OpenAPI 3.1 spec, with code samples in five languages — not an auto-dumped schema nobody read back.
- Hand-authored OpenAPI 3.1 spec
- Code samples in five languages
- llms.txt publishes the API map for AI assistants
- Hosted at app.apimio.com/developers
MCP server on npm, running against production
One npx line connects Claude — or any MCP client — to a merchant catalog, with 47 tools generated live from the OpenAPI spec, so new API capability reaches agent tooling without a separate release. The server is pointed at production: the tools operate on the real catalog under the permissions of the API key you give it.
- Published as @apimio/mcp on npm
- 47 tools generated from the live spec
- Runs against production, scoped by your API key
- Works with any MCP client; tools track the API automatically
Agent-ready feed format
A feed template shaped for LLM and agent consumption rather than for an advertising channel: complete attributes, resolved categories, plain-language descriptions, none of the channel-specific fields. Generated by the same feed engine as Google Shopping and Meta and served as a hosted URL — the catalog an AI shopping assistant can actually read. It can be mapped to what ChatGPT and OpenAI product-feed programs ask for; it is not an OpenAI specification (Advanced and Enterprise plans).
- Complete attributes and resolved categories
- Plain-language descriptions, no channel cruft
- Hosted URL, refreshed on the feed schedule
- Mappable to ChatGPT/OpenAI product-feed requirements
API writes are audited too
Changes made through the API or the MCP server are attributed in the same audit trail as changes made in the interface, so an automated write is exactly as traceable as a human one.
- API and MCP changes attributed in the audit trail
- Same history as interface edits
- Who, what, when, and the previous value
The API reads the same catalog everything else does
There is no separate developer data model to learn or keep in sync.
Catalog Hub
The canonical record the API reads and writes, synced two-way with your Shopify stores in real time.
Explore Catalog HubQuality Guard
Products written through the API are scored by the same rules, so an automated pipeline cannot quietly bypass your quality bar.
See Quality GuardProduct Feeds
Data pushed through the API flows into the same hosted channel feeds — Google, Meta, Pinterest, TikTok, Bing — without building a second integration (Advanced and Enterprise plans).
See Product FeedsChatGPT product feed
What an agent-ready product catalog is, how the feed format maps to what ChatGPT and OpenAI ask for, and how to serve it from Apimio.
See the agent-ready feedThe Apimio API — common questions
Yes — a public REST API v1 covering products, variants, media, categories, brands, vendors, attributes, inventory across all eight states, and multistore overrides and locales, for both reads and writes. It uses cursor pagination, accepts idempotency keys on writes, and offers bulk endpoints of up to 500 products per call with job-status polling. Apimio publishes the full reference at app.apimio.com/developers, and the API is on Advanced PIM and Enterprise PIM.
No. The public API is REST only. Apimio uses Shopify's GraphQL Admin API internally for sync, but what is exposed to you is REST — and it is better to say that plainly than let a GraphQL expectation form and disappoint someone mid-build.
Not yet. What exists is code samples in five languages in the reference, plus the OpenAPI 3.1 spec itself, which most language toolchains can generate a client from. Packaged, published SDKs are not something Apimio ships today, and the samples should not be mistaken for one.
MCP is the protocol AI clients use to reach external tools. Apimio publishes @apimio/mcp on npm, so one npx line connects Claude or any other MCP client to a merchant catalog with 47 tools — generated live from the OpenAPI spec, which means new API capability reaches agent tooling without waiting for a separate SDK release. It runs against production: the tools read and, with a scoped key that allows it, write the real catalog. It is genuinely published and installable, not an announcement of intent.
A product catalog an LLM or AI shopping assistant can read without guessing: every attribute present, categories resolved to a standard scheme such as Google's product taxonomy, descriptions in plain language, and no channel-specific abbreviations or codes. Apimio produces this as an agent-ready feed format on the same feed engine as its Google Shopping and Meta feeds, served at a hosted URL that refreshes on schedule (Advanced and Enterprise plans). It can be mapped to what the ChatGPT and OpenAI product-feed programs ask for; it is not an official OpenAI specification. The API and MCP server give agents the same catalog interactively.
Yes — outbound webhooks for product and variant events, HMAC-SHA256 signed so you can verify the payload came from Apimio, with automatic retries and backoff, an inspectable delivery log and manual redelivery. What is not supported is inbound public webhooks: you cannot register an endpoint that writes into Apimio. Writes go through the REST API.
Monthly call quotas run to 250,000 on Advanced PIM and 1,000,000 on Enterprise PIM, with rate limits scaled to the plan. Keys are org-bound and managed under role-based permissions, and an admin can grant access to a partner agency working on a client account. Apimio does not offer API access on Basic PIM.
Advanced PIM and Enterprise PIM, or by arrangement. It is not part of Basic PIM. If you are an agency needing programmatic access to a client account, an admin on that account can grant it to you. The MCP server and the agent-ready feed follow the same plan gating in Apimio (Advanced and Enterprise plans).
Read the spec before you commit to anything
The full OpenAPI reference, code samples and the MCP quickstart are in the developer portal, and nothing to sign up for to read them. When you want to build against a real catalog, start the 14-day free trial.