deepdive

Circle Discovery API: AI Agents Find and Pay for Services

Editorial · Aug 6, 2026 · 8 min read

Circle launched its Discovery API as the newest component of the Agent Stack, exposing a programmatic interface that lets AI agents search for USDC-compatible services using 14 query filters. The API moves agent commerce one step earlier in the transaction pipeline: instead of an agent only executing a payment once a service endpoint is already known, the agent can now discover, compare, and select services autonomously before spending a single cent of stablecoin. The launch signals that Circle is building vertically, constructing the full stack from discovery through settlement on rails it controls.

How the Discovery API Works

The Discovery API functions as a queryable directory of services that accept USDC, designed specifically for programmatic access by autonomous agents rather than human operators. An agent sends a structured request specifying what it needs, and the API returns matching service providers along with metadata that lets the agent evaluate suitability. The 14 query filters cover dimensions like service category, pricing model, geographic availability, latency requirements, supported chain, and provider reputation signals. The design assumes that agents need machine-readable service descriptions, not scraped web pages or human-negotiated contracts. An agent looking for image generation, translation, or data retrieval can query the directory, compare options on price and capability, and select a provider, all within a single logical loop.

Where Discovery Sits in the Agent Payment Stack

The Discovery API is not a standalone product. It is the front layer of Circle’s Agent Stack, which already includes USDC payment execution and programmable wallet infrastructure. The stack now spans three logical stages: discovery, evaluation, and settlement. An agent first queries the Discovery API to find candidate services. It then evaluates the results based on its own logic, cost constraints, or task requirements. Finally, it pays the selected provider in USDC through the existing payment layer. By owning the directory layer, Circle positions itself as the default routing point for agent commerce, similar to how traditional payment networks derive leverage from controlling both merchant onboarding and transaction processing.

The Competitive Positioning

Circle’s approach differs from other agent-commerce architectures we have tracked. Platforms like Skyfire and Payman build agent payment functionality on top of existing stablecoin rails but rely on external discovery mechanisms, whether through marketplace listings, manual configuration, or agent-to-agent negotiation. Coinbase’s Agent Payments focuses on the wallet and transaction layer through its MPC infrastructure. Cloudflare’s Monetization Gateway handles per-request payment at the edge using HTTP 402 but presumes the agent already knows which endpoint to call. Circle’s bet is that discovery is the bottleneck, not payment execution. If agents cannot find services programmatically, the volume of agent-driven USDC transactions stays trapped at low levels, which is exactly what Morgan Stanley flagged when it noted daily agentic payments were barely four figures.

Open Questions and Tradeoffs

The Discovery API solves a real gap, but several questions remain unresolved. Provider listings in the directory are presumably self-registered, which raises the question of quality control and how agents should weight reputation signals in a nascent ecosystem where few providers have transaction history. Pricing volatility across services is another concern: if two providers offer the same capability at different USDC prices, agents need reliable comparison logic, and the API’s metadata may not capture latency or reliability differences that matter in production. Finally, there is the fragmentation risk. If competing issuers build their own discovery layers, agents may need to query multiple directories, recreating the exact fragmentation problem the API set out to solve.

Sources

E
Editorial
Related reading

Related reading