Coinbase links AI and blockchain with Payments MCP, its new smart payment system

Última actualización: 10/24/2025
  • Payments MCP connects AI agents to on-chain value transfer via the Model Context Protocol and Coinbase’s x402 open payment standard.
  • Works with Anthropic’s Claude (Desktop and Code), Google Gemini, OpenAI Codex and Cherry Studio; ChatGPT support is planned.
  • No-code onboarding: email-based wallet creation, integrated onramp/guest checkout, local execution, and the x402 Bazaar Explorer for in-app paid APIs.
  • Safeguards include per-agent wallets, spend caps, approval thresholds, and KYC at on/off-ramps; x402 Foundation (with Cloudflare) keeps the spec vendor-neutral.

AI and blockchain payments illustration

Coinbase is tying artificial intelligence directly into crypto payments with a new stack called Payments MCP, designed to let AI agents create wallets, move funds in stablecoins, and interact with online services without human babysitting. By wiring agent capabilities to on-chain rails through the Model Context Protocol (MCP), the company is pitching a standardized way for software to transact value at scale.

Coinbase Payments MCP architecture

For hands-on trials, Coinbase’s developer platform provides an integrated interface where users can spin up and fund wallets with just an email address. There’s no developer setup or API keys required to start, and the x402 Bazaar Explorer lists APIs and services that agents can pay for inside the app. An embedded onramp and guest checkout let supported regions transact almost immediately.

Controls are baked in: users can set spending limits, approval thresholds, and session caps from a straightforward settings panel. For speed and privacy, Coinbase says the system executes locally, reducing back-and-forth and keeping sensitive agent activity closer to the user’s machine.

Coinbase frames the release as a bridge from developer tooling to everyday use, making “agentic commerce” accessible beyond programmers. In short, agents aren’t just processing information anymore — they’re now able to take direct economic actions within guardrails.

  Tokenization goes mainstream: indexes, real-world assets and payment rails converge

While Payments MCP already supports Claude, Gemini, Codex and Cherry Studio, ChatGPT isn’t included yet due to transport incompatibilities with OpenAI’s current streaming approach. Coinbase says support is on the roadmap.

How Payments MCP works

For hands-on trials, Coinbase’s developer platform provides an integrated interface where users can spin up and fund wallets with just an email address. There’s no developer setup or API keys required to start, and the x402 Bazaar Explorer lists APIs and services that agents can pay for inside the app. An embedded onramp and guest checkout let supported regions transact almost immediately.

Controls are baked in: users can set spending limits, approval thresholds, and session caps from a straightforward settings panel. For speed and privacy, Coinbase says the system executes locally, reducing back-and-forth and keeping sensitive agent activity closer to the user’s machine.

Coinbase frames the release as a bridge from developer tooling to everyday use, making “agentic commerce” accessible beyond programmers. In short, agents aren’t just processing information anymore — they’re now able to take direct economic actions within guardrails.

Under the hood, Payments MCP rides on x402 — Coinbase’s open payment spec inspired by the rarely used “HTTP 402: Payment Required” code. It currently plugs into Anthropic’s Claude (Desktop and Code), Google’s Gemini, OpenAI’s Codex, and Cherry Studio, giving agents a uniform interface to hold wallets, fund them, and send stablecoin payments straight from a prompt.

Why crypto fits machines

Erik Reppel, engineering director for Coinbase’s developer platform, argues that cryptocurrency is uniquely suited to software-driven payments because it is digital-native and open. In his view, agents are simply smarter programs that prefer programmatic interfaces over human-optimized checkout flows.

  JPMorgan Will Accept Bitcoin and Ether as Collateral for Institutional Loans

Reppel also explained the backstory of x402. The naming nods to long-running efforts to create an internet-native payment standard. Since the 1990s, teams (including Marc Andreessen’s at Netscape) experimented with web-based payments; HTTPS made card use on the web safer, but a single protocol never emerged. Today, merchants still handle card data differently, and behind familiar forms lie many bespoke computer-to-computer processes — not a true standard.

Security, safeguards, and compliance

Each agent operates with its own dedicated wallet, separate from a user’s primary holdings. Coinbase emphasizes configurable funding caps, approval thresholds, and session limits so that spending is bounded and visible.

Developers can layer rule engines on top of x402. For example, an agent might be allowed to spend up to ten cents autonomously but require approval for anything larger. The idea is that the money itself is programmable, so safeguards become more flexible and fine-grained.

On compliance, Coinbase says its framework mirrors these technical controls with checks at every layer, handling KYC at entry and exit points so it doesn’t derail the user experience. The company positions this as part of the value of using its developer platform: building compliance directly into the product surface where possible.

Open standard and adoption path

To keep the specification neutral, Coinbase and Cloudflare are backing the x402 Foundation. The goal is to ensure builders can rely on an open protocol that remains usable independent of any single company’s fate.

Looking ahead, Reppel expects 2026 to be a pivotal moment—the potential “year of agentic payments”—as AI systems begin buying compute, data, and other services programmatically. Most people, he suggests, won’t notice crypto under the hood: they’ll just see an AI balance tick down by a few dollars while stablecoins settle instantly behind the scenes.

  Bullish’s IPO Settled $1.15B in Stablecoins, Led by USDC on Solana

The launch also fits a wider arc tying AI to on-chain finance. Recent moves like Talos’ AI-managed treasury highlight a growing overlap between automation and DeFi. Coinbase’s play is to make stablecoins the default payment channel for autonomous agents, with x402 and MCP offering common rails for both developers and mainstream users.

Stepping back, Payments MCP binds AI agents to a standardized, on-chain payment layer using MCP plus the x402 spec, adds local-first privacy, and wraps it with spend controls and compliance checks. If it gains traction across model providers and tooling, agent-driven transactions could feel routine for end users while the heavy lifting happens with stablecoins under the surface.

Tether AI platform to support Bitcoin-1
Artículo relacionado:
Tether AI Platform Set to Bring Bitcoin and USDT Payments to Peer-to-Peer AI Ecosystem