What Agent Central does
Agent Central gives MCP-compatible AI clients access to Amazon Seller Central and Amazon Ads through one hosted connection. It is intended for sellers, operators, and agencies that want an assistant to work with advertising and commerce data in the same workflow.
The service covers Amazon Ads, including Sponsored Products, Sponsored Brands, Sponsored Display, DSP, and Stores. Seller Central coverage includes sales and traffic, orders, inventory, inbound shipments, catalog, fees, settlements, and fulfillment. This combination supports cross-domain questions such as comparing advertising spend with sales, calculating profit after fees and ad spend, and reviewing product-level performance.
Access can span 23 Amazon marketplaces, including the US, UK, Ireland, Germany, France, Italy, Spain, the Netherlands, Belgium, Sweden, and Poland. Agency users can separate client accounts into workspaces with distinct Amazon connections and keys.
How it works
The hosted MCP endpoint is the production service. Amazon account connections use Amazon’s own authorization pages, and the service accesses account data through the Selling Partner API and Amazon Ads API rather than scraping. After connecting accounts, an administrator creates an API key and adds the endpoint to an MCP client.
Keys may be read-only, restricted to areas such as advertising or inventory, or limited to selected tools. Write-capable workflows use a preview-first process: a change tool returns the proposed operation and relevant current values, and the change is not sent to Amazon until an explicit submission includes an idempotency key. Submitted changes are recorded with old and new values when existing settings are edited.
Agent Central returns Amazon data, source fields, deterministic metrics with stated formulas, and audited write results. The connected assistant remains responsible for interpreting those results and deciding what action to take.
Setup and configuration
Create an account at agentcentral.to, connect Seller Central and Amazon Ads, and create a key in the dashboard. A trial is available without a payment card. For Claude, paste the key’s Connector URL into a custom connector. Claude Code can use the hosted HTTP transport with AGENT_CENTRAL_API_KEY as a bearer token:
export AGENT_CENTRAL_API_KEY="ac_live_..."
claude mcp add --transport http agentcentral https://mcp.agentcentral.to/mcp \
--header 'Authorization: Bearer ${AGENT_CENTRAL_API_KEY}'
Other MCP clients can use the endpoint with an Authorization: Bearer header. ChatGPT web setup uses Apps, Advanced Settings, Developer Mode, and Create App; read-only access can also use OAuth at the hosted endpoint. Keep API keys and Connector URLs out of public repositories, chats, and issues.
Tools and capabilities
The hosted endpoint exposes 177 domain-scoped tools and five utilities, for 182 tools total. Supported workflows include:
- Querying Amazon Ads performance and settings
- Reviewing sales, traffic, orders, inventory, and inbound shipments
- Inspecting catalog data, fees, settlements, and fulfillment records
- Comparing advertising and Seller Central history across periods
- Previewing and submitting supported account changes with audit records
- Restricting access by domain, individual tool, or read-only policy
The repository also contains a local stdio introspection stub. It lists the public tool catalog but does not execute calls; real operations use the hosted endpoint.
Limitations and notes
The local stdio stub is not the production service and is not published to npm. It can be built from the repository or run with Docker for catalog introspection, but tool calls return a pointer to the hosted endpoint and setup guide. A separate fixture-only Claude Code example uses synthetic records and cannot access Amazon or customer data.
Historical comparisons depend on account connection and synchronization history. Amazon Ads history is retained while the account remains connected, and synced Seller Central history is retained while the account is active. The assistant receives data and metrics; recommendations and reasoning are performed by the assistant rather than by Agent Central.