Agents to transact with: Introducing Amazon Bedrock AgentCore payments built with Coinbase and Stripe

Machine Learning


We are in the midst of a fundamental shift in how software is built and used. AI agents are more than assistants waiting for instructions. Call APIs on your behalf, access MCP servers, and coordinate with other agents to complete complex multi-step tasks. As the tasks agents perform become increasingly diverse, the ecosystem around them is expanding just as quickly to meet that demand.

Looking further ahead, services, tools, and content must be designed for humans and agents. Agents discover, evaluate, and pay for resources when they are needed, all within a single execution loop. The services that support them must be priced and used in a way that is billed in real time, in fractions of a penny per call. Early protocols like x402, ACP, MPP, and AP2 pioneered what this could look like, and teams are experimenting with payment-enabled agents. Although the building blocks are emerging, the agent economy is still in its infancy and the infrastructure to support it at scale does not yet exist. For developers who want to get ahead of the curve, the road has been difficult. Create bespoke billing relationships with all your service providers, securely manage credentials, strengthen spend governance, manage compliance requirements, and create orchestration logic across fragmented landscapes. This requires months of engineering effort and the stakes are high. A misconfigured payment flow will not only generate bad answers, but will also result in real money being moved.

Today, we are announcing Amazon Bedrock AgentCore Payments (Preview), a new feature set for Amazon Bedrock AgentCore. This allows AI agents to instantly access and pay for anything they use, including web content, APIs, MCP servers, and other agents. These features were built in partnership with Coinbase and Stripe, who provide the wallet infrastructure and payment rails that power the initial feature set.

AgentCore is a platform for building, connecting, and optimizing agents at scale, with enhanced security at the infrastructure layer that agents cannot bypass. Developers at companies like Cox Automotive, Thomson Reuters, and PGA TOUR are already using AgentCore to build agents that reason, plan, and execute across complex workflows. With today’s announcement, these agents will be able to conduct transactions using the same identity systems, agent gateways, and observability they already rely on. Amazon Bedrock AgentCore Payment is not an additional module; it is native to the platform on which the agent is built, and is governed by the same controls as all other actions the agent performs.

The first managed end-to-end payments feature for agents

It is the first managed payments capability built specifically for autonomous agents, covering the entire lifecycle from wallet authentication to transaction execution to spend governance and observability, allowing developers to focus on how their agents operate rather than how they pay.

These features allow developers to build agents that can access the resources they need, paid or unpaid, without having to manually tie together each billing relationship. Financial research agents can dynamically access real-time market data feeds and paywalled publications and pay for the articles and data points they use on behalf of end users. Coding agents can call specialized APIs or paid MCP servers as needed, such as private package registries, sandbox execution environments, or niche third-party agents that do one thing well. As the market matures, agents will be able to handle commercial transactions. This means they can book flights, book hotels, and complete purchases on your behalf across merchant platforms.

First, developers connect their agents to their wallets or payment service providers, register funded payment sources, and set per-session spending limits. AgentCore manages all credential authentication and token lifecycles. When an agent encounters a paid resource during execution, AgentCore handles protocol negotiation, retries, payments, and routes the transaction through the appropriate provider without interrupting the agent’s inference loop. All transactions can be monitored through the same logs, metrics, and traces that developers already use to monitor agent behavior.

AgentCore is designed to work with any framework and any protocol. We now bring that same flexibility to Amazon Bedrock AgentCore payments. Developers don’t have to keep track of the evolving landscape of payment protocols or stick to a single standard. At the time of preview, we support the x402 protocol, with additional protocols on the roadmap. As new protocols emerge, support is added at the platform level so that developers do not have to rebuild their agents.

“At Warner Bros. Discovery, we are actively exploring more flexible and scalable payment approaches as we evolve beyond direct API integration with third-party processors. AgentCore payments represents a promising direction and allows teams to experiment with agent-driven experiences, allowing premium content like live sports and featured releases to be seamlessly displayed and transacted at the moment of interest. We are particularly interested in evaluating the potential for reducing overhead, streamlining payment orchestration, and implementing governed services to enable traceable transactions as we consider potential next-generation commerce experiences.” – Mit Majithia, Executive Vice President, Warner Bros. Discovery.

Micropayments in preview: Unlock paid data, APIs, and content for agent workflows

The first use case we’ll enable in the preview is for agents to make instant micropayments to access APIs, MCP servers, web content, and other agents. Services are rapidly moving to a pay-as-you-go model, and AI agent web crawling has skyrocketed over the past year, with these transactions typically costing less than a dollar or a few cents.

Developers enable Amazon Bedrock AgentCore payments on existing agents using the AgentCore SDK or the console. Select either Coinbase wallet or Stripe Privy wallet as your payment connection. Both options allow end users to fund their wallets through stablecoins or fiat currency using a debit card. Guardrails will be enforced in multiple layers. Before an agent can perform a transaction, the end user must explicitly authorize the agent to access and use their wallet. At runtime, spending limits are applied on a per-session basis to keep the agent within the set budget for each run. Agents never have unlimited access to funds. Operates only with explicit permission and within defined limits.

Under the hood, the payment flow is built around the x402 protocol, an open HTTP native payment standard that enables instant stablecoin micropayments. When the agent sends a request to the paid endpoint and receives an HTTP 402 “Payment Required” response, the payment process authenticates with the configured wallet, performs the stablecoin payment, attaches proof of payment, and sends the content back to the agent. All this happens within a run loop. Payment managers orchestrate flow, and payment limits track overall spending against session budgets. Once enabled, agents will begin reconciling payments on the fly, with full traceability available in the AgentCore console.

To enable agents to discover merchants on their own, we make the Coinbase x402 Bazaar MCP server available through the AgentCore gateway. Bazaar provides a curated list of x402 endpoints that agents can search, discover, and pay for when relevant to their tasks, turning paid services into something agents can discover and use themselves, rather than having developers hard-code each integration.

Heurist AI, a full-stack infrastructure provider for the AI ​​economy, uses AgentCore’s payments capabilities to build research agents that perform financial analysis on behalf of end users. “Heurist uses AgentCore payments as a research agent to help end customers perform financial and cryptocurrency analysis and investment advice,” said JW Wang, founder of Heurist AI. “End customers can set research budgets, and agents use AgentCore payments to get accurate, real-time data about the market, social sentiment, and news in general. We were able to quickly integrate agent payments with little effort and a few lines of code.”

Built with an ecosystem of partners

Coinbase developed the x402 protocol, an open standard rapidly gaining traction in machine-to-machine payments, and built the CDP wallet infrastructure and facilitation to power micropayment flows in preview. Coinbase has been innovating on AWS for years, serving millions of customers with its cryptocurrency exchange and developer platform. We are currently working together as members of the x402 Foundation to establish open standards for the agent economy.

“Soon, more AI agents will be transacting than humans, and they will need programmable, always-on, global funds built for the internet. By bringing Coinbase’s stablecoin infrastructure and x402 to AWS AgentCore, we give developers a full stack to build agents that move funds at the speed of software with the reliability and compliance that enterprises expect.” – Brian Foster, Head of Infrastructure Growth and Strategy, Coinbase.

Stripe is helping define how commerce works in the age of agents by building tools that enable AI agents to discover, negotiate, and complete transactions on behalf of businesses and consumers. With the launch of the new payments feature, AgentCore will integrate Stripe’s wallet infrastructure powered by Privy as a payments connection in preview, giving developers direct access to Stripe’s payments infrastructure from day one. Together, AWS and Stripe are working toward a common path to fiat payments support, expanding beyond micropayments and combining Stripe’s global payments reach with the agent platform developers are already building.

“Stripe is building the economic infrastructure for AI. For agents to be meaningful economic actors, they need a way to hold and spend money. That’s why we’re excited to partner with AWS to make a stablecoin wallet for agents immediately available to AgentCore developers.” – Henri Stern, CEO of Privy, Stripe.

where we are heading

Micropayments are the first step, addressing the earliest agent-to-agent commerce patterns where the most immediate appeal is seen. Beyond micropayments, we see a natural expansion into broader commerce flows where agents act on behalf of buyers as well as other agents. Agents who book flights, book hotels, and complete purchases across sales platforms on behalf of customers. Getting there will require deeper integration with the payments ecosystem, support for additional protocols, stronger buyer intent validation, and end-to-end observability across the transaction lifecycle. That’s the way forward and we’re building towards it.

The developer experience is consistent across each phase. Configure wallets, set policies, and conduct agent transactions. What changes is the range of how much and how agents can be paid. I can’t wait to see what happens next.

Let’s get started

Amazon Bedrock AgentCore Payments is currently available in preview in US East (N. Virginia), US West (Oregon), Europe (Frankfurt), and Asia Pacific (Sydney). Start with the AgentCore console. See the documentation for more information.


About the author

Pretty CN

Pretty CN

Preethi is a Director at AgentCore, an Agentic AI organization, with over 20 years of expertise in embedded and cloud software development. In her 14 years at Amazon, she designed large-scale distributed systems, drove AI innovation across Retail, Alexa, and AWS, and delivered breakthroughs in multimodal AI. She led voice recognition in Alexa, computer vision services in AWS, and generative AI transformations that revolutionized how organizations extract insights from unstructured content at scale. As a technical advisor for the Agentic AI Organization, she provided strategic oversight across Amazon Quick, Kiro, and AWS Transform. Most recently, she created the vision and led the launch of AgentCore, a platform for building, connecting, and optimizing production-ready AI agents at scale.



Source link