Going Deeper
Scope Your API Keys
Give each agent, person, or marketplace the smallest useful tool surface instead of sharing one broad key.
You don't have to give your agent access to everything
The idea of an AI agent with full access to your Amazon account can feel risky. That's why every agentcentral key can be set to read-only, which means your agent can pull data across all the tools it has access to but can't modify anything in Amazon. For a lot of sellers, especially early on, that's exactly the right setup: you get the analysis and reporting without any risk that a bad prompt accidentally changes a live campaign.
Once you're comfortable, you can create a separate key with write access for specific workflows. But there's no rush. Read-only is a perfectly good default.
Beyond that, there are three dimensions to think about as you set up your keys.
Scope by workflow. agentcentral has 141 tools across ads, inventory, catalog, finance, ranking, and fulfillment. When your agent sees all of them, it has to decide which ones to call based on your question, and a broader tool surface means more room for it to pick the wrong one. A key scoped to just the ads read tools means your agent only sees campaign performance, search terms, keyword data, and related tools. It can't accidentally pull inventory data when you asked about ad spend.
Name the workflow before you create the key. If you can't name it, the key is probably too broad. A daily sales reviewer probably needs sales, orders, listings, and inventory read tools, but not ads writes or catalog writes. An ads analyst needs campaign performance, search terms, keyword performance, and TACOS, but shouldn't have bid and budget write access unless that's part of an approved workflow.
Scope by person or agent. Your inventory manager gets a key with stock levels, sales velocity, inbound shipments, and days of cover, but no ads tools and no write access. Your PPC agent gets a dedicated key scoped to just the ads read tools, so it only sees campaign performance, search terms, keyword data, and TACOS. A read-only weekly review agent can see everything but change nothing.
Each one gets its own key scoped to its job. When an agent only has the tools it needs, it picks the right one faster and can't accidentally wander into the wrong domain. When someone's role changes or you retire a workflow, you revoke one key and everything else keeps running.
Scope by marketplace. Each agentcentral API key is scoped to one marketplace. If you sell on both the US and Canadian marketplaces, you need two keys. Include the marketplace in the key label: “ads-read-CA,” “inventory-US,” “full-suite-DE.” When your agent has multiple keys, clear naming prevents it from querying the wrong marketplace. You can give one agent multiple marketplace keys for cross-marketplace comparisons, or run separate agents per marketplace if you want isolated workflows (which is usually cleaner for scheduled reports).
You can create and scope keys from the dashboard. Each key gets a label, a set of domains, and optionally a list of specific tools within those domains. Start with read-only, name your workflows, and only add write tools when you've decided you trust a specific workflow to make changes. You can always create a new key — you're never locked into a decision you made during onboarding.
Related Reference
Continue Reading
Next: Build A Prompt LibraryTurn useful conversations into reusable one-shot prompts with precise dates, metrics, and output formats.
Related Best Practices
Run the account data tour to see what data is available, what is syncing, and what needs attention.
Submit feedbackTell agentcentral what worked and what did not, directly from your agent conversation.
Prompt libraryTurn useful conversations into reusable one-shot prompts with precise dates, metrics, and output formats.
Business playbookTeach your agent your ACOS targets, lead times, seasonal patterns, and preferred output formats so every conversation starts informed.