The useful distinction here is that writes are a different class of tool with previews and audit logs behind them, and a key can be issued read-only — so you can let an assistant answer questions about spend and stock cover without also granting it the ability to change a price. Everything real happens on the hosted endpoint; cloning the repository gets you a catalog listing and nothing else.
A hosted MCP server for Amazon sellers, reached over Streamable HTTP at `https://mcp.agentcentral.to/mcp`. It connects to Amazon Ads and Seller Central through Amazon OAuth and returns factual seller data with source fields — the repository itself holds registry metadata and a stdio stub for introspection, not the working server.
- Amazon Ads reporting: campaigns, ad groups, keywords, targets, search terms, placements, budget pacing, TACOS, DSP, Brand Store and Amazon Attribution
- Seller Central operations: inventory, orders, returns, reimbursements, listings, suppressed listings, inbound shipments, and FBA/AWD stock facts
- Catalog and listing quality: details, sales ranks, A+ Content status, variations, listing issues, reviews and keyword ranks
- Money: finance, profitability, payment transactions, settlements, fulfillment, MCF shipping and MCF orders
- Guarded write tools for supported Ads, catalog, listing, price, inventory-quantity and MCF operations, with previews, guardrails and audited results
- The hosted endpoint exposes 141 domain-scoped tools and 145 production tools in total
An agentcentral account, your Amazon Ads and Seller Central connected over Amazon OAuth, and a scoped API key. Add the endpoint to any HTTP-capable MCP client with an `Authorization: Bearer ac_live_<YOUR_API_KEY>` header; Claude custom connectors instead use a signed connector URL generated in the dashboard. Keys can be scoped read-only. Nothing to install — the repository's stdio stub only lists the catalog and never executes a call.
One command plus a key — npx -y mcp-remote https://mcp.agentcentral.to/mcp, then supply credentials