Building a small x402 API surface for autonomous agents
I wanted to test a simple idea: can a few narrow, deterministic utilities be easier for an agent to buy than a large all-purpose API?
The result is HustleOS Agent Utilities, a small public x402 service on Base. Each endpoint returns a normal HTTP 402 challenge, includes a Bazaar discovery declaration, and charges only on a successful request.
The five utilities
- Landing-page evidence ā source-visible metadata, headings, action labels, forms, and JSON-LD presence for one public page.
-
Agent-discovery readiness ā checks the observable presence and basic structure of
robots.txt,llms.txt,/.well-known/agent.json, andopenapi.json. - Catalog gateway health ā a bounded public PayanAgent gateway-reachability snapshot without invoking paid tools.
- Agent work radar ā policy-filtered public paid-work leads across several agent marketplaces.
- TaskMarket candidates ā a lower-cost filtered TaskMarket feed.
Design constraints
The tools deliberately avoid credentials, wallet access, private data, web logins, and side effects. The public-site utilities accept HTTPS only and document their limits instead of claiming conversion, ranking, security, or third-party quality.
The service catalog is machine-readable:
I would value technical feedback on the 402 discovery data and on which small, verifiable agent utilities are actually worth paying for.
Offline developer utilities
For people who prefer a local command-line workflow, I also packaged two small downloads:
- Endpoint Health Checker ā Agent Utility ($3): public catalog gateway-reachability checks with no wallet, payment, or credentials.
- Agent Discovery Readiness Audit ($5): a bounded, local check of common public discovery files.
Both are deliberately narrow, use public data only, and do not guarantee third-party quality, safety, or compatibility.
Top comments (1)
The narrow-scope design is the right instinct. For agent-facing APIs, Iād rather see small tools with explicit inputs, observable outputs, and clear limits than a large endpoint that tries to infer too much.
One thing Iād add is a confidence or evidence field in the response schema. If an agent is using the result inside a workflow, it helps to know whether the output came from direct page evidence, a missing file, a partial crawl, or a derived assumption. That makes downstream decisions easier to audit.