product updateAmazon Web Services

AWS Launches Bedrock AgentCore Payments to General Availability, Letting AI Agents Pay Autonomously

TL;DR

Amazon Bedrock AgentCore Payments, built with Coinbase and Stripe, is now generally available after a May preview. The service lets AI agents autonomously pay for paywalled content, APIs, and per-inference compute using stablecoin wallets with spending caps and audit logging.

3 min read
0

Amazon Web Services announced general availability of Bedrock AgentCore Payments, a service that lets AI agents autonomously pay for APIs, MCP servers, and paywalled web content. The service moved from preview, launched in May in collaboration with Coinbase and Stripe, to full production availability.

AgentCore Payments addresses a specific gap: AI agents can reason and select tools, but most lack a way to actually pay for metered services when a task requires it. AWS's fix connects agents to stablecoin wallets — via Coinbase and Stripe Privy — that support microtransactions often priced in cents. End users fund these wallets through credit cards or USDC and grant delegated spending authority to their agents.

Security model. Developer credentials for wallet providers are stored in AgentCore Identity Secrets Manager. Agents never see raw credentials; instead, AgentCore issues short-lived tokens derived from those credentials to authorize wallet operations like transaction signing.

Protocol support. At preview, AgentCore supported only the x402 protocol. At GA, AWS added support for the Machine Payment Protocol (MPP), co-authored by Stripe and Tempo, letting developers pay MPP-compatible services without extra code. AWS also added an "upto" scheme within x402 that lets agents set a spending ceiling rather than commit to a fixed price — enabling usage-metered billing for things like LLM tokens or compute, versus the fixed-price "exact" scheme available since preview.

Spending controls. Transactions run inside a "payment session" — a scoped context for a single agent interaction with two configurable limits: a maximum spend amount and an expiry time. AWS says these checks run deterministically at the infrastructure layer, rejecting any request that would exceed the session budget, a safeguard against non-deterministic agent behavior such as misread authorizations or duplicate retries.

Observability. AgentCore Payments integrates with AgentCore Observability, emitting logs to Amazon CloudWatch and spans for prebuilt dashboards tracking transaction success rates, average transaction values, and spending behavior across agents and sessions.

Discoverability. AWS exposes Coinbase's pay-per-use x402 endpoints as an MCP server through AgentCore Gateway, now curated by social proof, metadata quality, and availability, according to AWS.

AWS named early adopters spanning several use cases. Anchor Browser, a cloud browser automation platform, uses the service to unlock paywalled web content for agent workflows, according to CEO Idan Raman. Cloudflare has integrated its Monetization Gateway so content and API owners can charge agents directly via x402. SpreadX uses AgentCore Payments in its Incarna product to pay for inference through BlockRun's routing layer. Travala has integrated payments into its Travel MCP servers for conversational hotel booking through platforms like Claude, and Elsa AI and Heurist AI use the service for pay-per-use financial research APIs.

AWS did not disclose pricing for the AgentCore Payments service itself.

What this means: This is AWS formalizing infrastructure for the "agent economy" — a bet that AI agents will increasingly transact directly with services rather than relying on human-managed API keys and subscriptions. The technical bar here is trust: deterministic spending caps, secrets isolation, and audit trails are the mechanisms that let enterprises actually deploy autonomous payment in production rather than sandbox demos. The move to support the "upto" scheme is notable because it enables true metered pricing (pay for exactly what you use) rather than forcing providers into flat per-call pricing — a meaningful shift for anyone selling inference or compute by the token. Competitors including Google and Microsoft will likely need comparable payment rails for their own agent platforms, and the involvement of Stripe, Coinbase, and Cloudflare suggests x402 and MPP are becoming the de facto standards agents will use to pay each other and third-party services.

Related Articles

product update

AWS adds managed Web Search to Claude Desktop via Bedrock AgentCore Gateway in three Regions

AWS published a walkthrough for connecting Claude Desktop on Amazon Bedrock to a managed, MCP-compatible Web Search capability through Amazon Bedrock AgentCore Gateway. According to AWS, the search is backed by an Amazon web index spanning tens of billions of documents, and query traffic stays within AWS infrastructure. Web Search is available in three AWS Regions; pricing is not disclosed in the post.

model release

Amazon open-sources Strands Decider 2B, a small decision model built on a Qwen3.5-2B base

Amazon Web Services has released Strands Decider 2B, an open-source model that chooses among pre-decided options and returns a confidence score instead of generating text. It is inspired by TypeSafe's Jev and is small enough to run locally. Amazon says it briefly topped the Jevbench ranking for models of its size.

product update

Pi 1.0 agent harness goes stable; Pi Durable ports it to TypeScript with crash-resumable state

Pi, the minimalist agent harness now under Earendil, has reached version 1.0. A companion release, Pi Durable, ports it to TypeScript and externalizes all stateful components so agents can resume after crashes. Pricing and licensing were not disclosed in the source.

product update

AWS details ambient agent pattern on Bedrock AgentCore: S3 events trigger jobs, one ask_human tool pauses for approval

AWS published a reference implementation for ambient agents on Amazon Bedrock AgentCore. S3 uploads or scheduled events create jobs that an agent runs, pausing for human input through a single ask_human tool. Each agent turn is capped at the 15-minute Lambda timeout.

Comments

Loading...