One measured layer between your Amazon account and your AI

AgentXray reads the settlement lines Amazon actually posted, prices every advertising arm against the seller's own fees, and hands the result to the assistant they already use over the Model Context Protocol: 32 tools, 19 that read measured numbers with their basis and gaps, 13 that file drafts which wait for a person. Walked end to end with Claude, Claude Code and ChatGPT; listed in the official MCP Registry; any MCP client speaks the same protocol.

Drafts, never actions

A connected assistant can file a bid, budget, negative keyword, campaign state, listing or shipment draft. The draft waits in an approval queue; the connector token cannot authenticate against the approval endpoint, so nothing reaches Amazon until a person approves it in the browser. Every applied change is judged again after fourteen days by net contribution.

19 read tools

• daily_brief
• did_my_changes_work
• get_ads_campaigns
• get_ads_daily_series
• get_ads_negatives
• get_ads_overview
• get_ads_placements
• get_ads_search_terms
• get_ads_targets
• get_amazon_bid_recommendations
• get_buybox_status
• get_inventory_age
• get_my_changes
• get_my_product_costs
• get_profit_by_sku
• get_profit_summary
• get_traffic_and_conversion
• list_my_marketplaces
• what_i_cannot_tell_you

13 draft tools

• draft_add_keywords
• draft_add_product_targets
• draft_bid_change
• draft_budget_change
• draft_campaign_state
• draft_create_campaign
• draft_create_listing
• draft_create_product_ad
• draft_image_swap
• draft_listing_update
• draft_negative_keyword
• draft_shipment_plan
• draft_target_state

Connect an assistant · How seller data is kept apart · Pricing