All 20 Criteria
P1-A Structured Data — 4/5
Homepage includes Organization, WebSite, SoftwareApplication, and FAQPage schema.org markup confirmed via scrape. Missing Offer and AggregateRating schemas that would enable richer agent inference about services and quality.
P1-B Machine-Readable Pricing — 3/5
Pricing presented in clear HTML tables on /pricing: per-call from $0.001, commission tiers (0%/5%/10%), health-score discounts (6–10%). Structured and readable but not encoded in schema.org/Offer or JSON-LD format. Agent must parse prose HTML.
P1-C llms.txt / Agent Layer — 5/5
/llms.txt confirmed present and referenced explicitly in robots.txt. robots.txt explicitly allows AI crawlers by name: GPTBot, ChatGPT-User, Google-Extended, Anthropic-AI, PerplexityBot, ClaudeBot. This is textbook agent-forward signal architecture.
P1-D API / MCP Availability — 4/5
MCP Tool Descriptor generation is a core platform feature. Proxy Key infrastructure routes agent API calls. Public marketplace registry is agent-queryable. No explicitly published OpenAPI spec or MCP server card found in docs search.
P1-E Discoverability (GEO) — 4/5
robots.txt grants full crawl access to all major AI crawlers. Sitemap present. Agent-facing vocabulary throughout content. Small current footprint (16 services, 23 agents) limits LLM training exposure, preventing a 5.
P2-A Offer Completeness — 4/5
Homepage + pricing page together communicate what (AI service marketplace), who (agent builders and service providers), and how much (per-call from $0.001, 0–10% commission). Most of this is findable from a single page scrape.
P2-B Scope & Limits — 4/5
Rate limits explicitly stated in ToS: 60 req/min, 300 burst per client; providers can set stricter limits. Health score metrics (uptime, response time, error rate) defined. Commission caps explicitly bounded. Strong for an early-stage platform.
P2-C Substitution Rules — 1/5
No evidence of explicit substitution or fallback rules for service unavailability. If a listed service goes down, there is no documented agent-facing protocol for what happens next.
P2-D Conditional Logic — 3/5
Commission conditions are disclosed across pricing and terms pages (time-based tiers + quality tiers). 30-day notice for changes stated in ToS. Not fully machine-readable in one place; agent must cross-reference pages.
P2-E Semantic Precision — 4/5
Precise language throughout: health scores 0–100, commission percentages as exact figures, rate limits as specific numbers, USDC settlement on-chain. Minor marketing phrases ("AI agents automatically find it") present but do not obscure core offer data.
P3-A Verifiable Performance — 3/5
99.9% monthly SLA with tiered credit structure (10%/25%/50%) documented in ToS. Health scores calculated continuously over 30-day windows. No public third-party status page found. Self-reported; no G2, Trustpilot, or uptime.io verification discovered.
P3-B Scoped Permissions — 4/5
Proxy Key system prevents agents from ever accessing provider credentials directly. Per-provider rate limits allow fine-grained access control. Missing explicit time-bounded or dollar-bounded agent permission scoping, which would make this a 5.
P3-C Audit Trail — 2/5
On-chain USDC payments via Base L2 create an inherently verifiable per-transaction audit trail (blockchain = public ledger). ToS requires reporting of unauthorized access. No agent-accessible transaction log API or machine-readable audit endpoint documented.
P3-D Behavioral Consistency — 3/5
30-day advance notice for commission changes stated in ToS. Terms last updated March 2026 (recent). MIT-licensed open source codebase on GitHub provides stability evidence. No version-controlled ToS or published changelog found.
P4-A Friction-Free Activation — 4/5
"3-minute setup without coding" for providers. Agent buyers get single-integration access to all marketplace services immediately. USDC payments auto-settle. The portal/dashboard step creates a minor human gate for initial provider onboarding but agent consumption is frictionless.
P4-B Agent Decision Signals — 4/5
Health scores (0–100), per-call price in USDC, quality tier (Standard/Verified/Premium), category tags, active agent count, and service call volume all provide machine-legible signals for agent evaluation and selection. Richer than most competitors at this stage.
P5-A Integration Depth — 3/5
Single marketplace integration grants access to all services — agents that wire in once are unlikely to rebuild from scratch. MCP Tool Descriptor format creates mild switching cost. Network effects are nascent (16 services) and won't create deep lock-in until marketplace density increases.
P5-B Agent Memory Layer — 1/5
No agent memory or personalization layer detected. Each transaction appears stateless. No agent profile, preference history, or session context found.
P5-C Programmatic Renewal — 3/5
Pay-per-use USDC model is inherently agent-executable: an agent with a funded wallet auto-pays per call with no renewal friction. No subscription renewal API needed. Deducted points because wallet-funding for recurring use still requires human setup.
P5-D Compounding Value Signal — 2/5
Health scores accumulate over 30-day windows (a form of compounding track record). No agent-readable signal exposing how a provider's capabilities or quality improve with usage history. Static marketplace today.
Rubric v1 (April 2026). Scores reflect the company's state on the audit date and may have improved since.