Amazon has launched the open beta of its Model Context Protocol (MCP) Server, a system that acts as a universal translator for its ad platform, allowing AI agents to run campaigns via simple English prompts, as reported by Adweek. The move is designed to simplify the notoriously difficult process of managing ads on its sprawling platform.
Robots get confused: The new system is a direct response to Amazon's own internal headaches with AI. The company found its agents could be unreliable, sometimes taking bizarrely inefficient routes like analyzing three years of data for a simple report or defaulting to obsolete code to get a job done. The problem, dubbed 'reasoning overload,' is that agents struggle to navigate the thousands of granular API calls needed for basic advertising workflows.
The instruction manual: Instead of forcing AIs to learn the arcane mechanics of its ad system, Amazon is giving them an instruction manual. The MCP server bundles common advertising jobs into pre-packaged "tools," all built on the open-source Model Context Protocol originally developed by Anthropic. Now, complex operations that once required dozens of manual steps—like creating new ad groups or launching a multi-stage campaign—can be executed with a single conversational prompt.
The move extends the agentic capabilities Amazon first introduced at its unBoxed conference last year and aligns with CEO Andy Jassy's vision for agentic commerce. Backed by an ad division that grew to nearly $18 billion last quarter, Amazon is positioning itself to become the default operating system for the next generation of AI-driven commerce.
Amazon's move is part of a broader industry race to standardize how AI agents interact with ad systems, with similar efforts announced last year. It also reflects the ad industry's larger shift away from manual workflows and toward agent-led automation as a key theme for 2026.
