Arcade.dev Human-Dependent

Name: Agent Native Offer Audit — Arcade.dev
Item: Arcade.dev
Rating: 62
Author: Agent Native Offers

AUDIE Score: 62/100 · Audited 2026-04-08 · Website: https://www.arcade.dev · Machine-readable: JSON

Pillar Scores

P1 Signal Architecture — 16/25

P2 Clarity Stack — 15/25

P3 Trust Envelope — 13/20

P4 Velocity Triggers — 8/10

P5 Gravity Design — 10/20

Executive Summary

Arcade.dev sits at the top of the Human-Dependent tier (62/100), on the cusp of Emerging. Its technical foundations are genuinely strong — an OpenAPI 3.0 spec, published MCP server, 7,000+ tools, and semantically precise documentation make it one of the most agent-accessible platforms in its category. The biggest gaps are architectural rather than technical: no schema.org structured data on the website itself, no fallback protocol documentation for tool failures, and no persistent memory layer for agents. Arcade's single highest-ROI action is adding JSON-LD to its pricing page — a one-hour fix that would make its offer immediately machine-parseable by any LLM retrieval system.

Strongest Signals

OpenAPI 3.0 spec + MCP runtime (P1-D: 5/5): Arcade is the most developer-complete platform in this audit category — OpenAPI spec at a known URL, MCP server published, 7,000+ tools, multi-framework compatibility. Any AI agent can discover and interact with Arcade's infrastructure programmatically.
Semantic precision in all communications (P2-E: 5/5): No vague marketing language. Every claim is technically specific: "user-specific permissions vs. service accounts," "OAuth token management," "contextual access controls." An AI agent can reason about what Arcade does without disambiguating human-speak.
Public status page with 90-day uptime history (P3-A: 3/5): Real-time uptime monitoring for all components including third-party API dependencies. 100% uptime reported. Publicly accessible and machine-readable for agent health checks.
Weekly changelog, semantic versioning (P3-D: 3/5): Active development cadence with structured, dated release notes creates a behavioral consistency signal that agents and developers can monitor for breaking changes.

Critical Gaps

No schema.org structured data (P1-A: 2/5): Despite being an agent infrastructure company, Arcade's own website has no JSON-LD markup for its organization, products, or pricing. An LLM crawling the web for tool-calling platforms will find Arcade's content but not in a machine-native format.
No substitution/fallback rules (P2-C: 1/5): When a third-party tool (GitHub, Slack, Stripe) goes down — which Arcade's own status page monitors — there is no documented fallback protocol for agent workflows. This is a critical gap for production deployments.
No agent memory layer (P5-B: 2/5): Arcade positions itself as a multi-user agent runtime but has no persistent memory or personalization layer that agents can read/write to. Every session starts from scratch at the platform level.
Audit logs gated to Enterprise only (P3-C: 3/5): Machine-accessible transaction logs are critical for agent governance. Locking this to Enterprise with no agent-accessible log API for lower tiers creates a governance blind spot for most users.

Priority Actions

Add schema.org JSON-LD to homepage and pricing page — +4 pts · P1 · Effort: Low
Publish a documented fallback/degradation protocol for tool unavailability — +3 pts · P2 · Effort: Low
Add agent-accessible memory/context API — +4 pts · P5 · Effort: High
Extend audit log API to Growth tier — +2 pts · P3 · Effort: Medium
Publish programmatic usage/billing status endpoint — +2 pts · P4 · Effort: Low

All 20 Criteria

P1-A Structured Data — 2/5
No schema.org/JSON-LD markup found on the homepage or pricing page. The site is Next.js-based (modern React SPA), but no Product, Offer, or Organization schema was discoverable. robots.txt is permissive (`Allow: /`) with sitemap linked, but no rich structured data signals for AI retrieval.

P1-B Machine-Readable Pricing — 3/5
Pricing is presented in clean HTML tables with three tiers (Hobby $0, Growth $25/mo, Enterprise custom). Numeric values are readable in HTML (1,000 standard tool executions, $0.01/execution overage, etc.) but no schema.org/Offer or JSON pricing spec is published. Good clarity in prose; not yet machine-tagged.

P1-C llms.txt / Agent Identity Layer — 3/5
Arcade has an `/llms.txt` file served at `arcade.dev/llms.txt`. It contains structured platform documentation intended for LLM consumption (confirmed accessible). No `/llms-full.txt` found (404). Developer docs emphasize "LLM-friendly" documentation with clean markdown — above average but not a full agent identity layer.

P1-D API / MCP Availability — 5/5
Full OpenAPI 3.0 spec at `https://api.arcade.dev/v1/swagger`. MCP server published and documented. Supports MCP clients (Cursor, Claude Desktop, VS Code, Copilot Studio). Python and JavaScript SDKs available. Pre-built catalog of 7,000+ tools. This is the platform's core strength.

P1-E Discoverability (GEO) — 3/5
Active blog (State of AI Agents 2026, integration trend reports), clean sitemap, permissive crawl policy. Listed on G2 and Product Hunt. However, no AI-retrieval optimization layer (no semantic markup, no structured FAQs). Reasonable organic presence for the category.

P2-A Offer Completeness — 3/5
Pricing page clearly states what each tier includes and costs. Homepage describes the product in concrete terms ("MCP runtime," "authenticated tool-calling platform"). What is slightly unclear: the relationship between "user challenges," "standard tool executions," and "pro executions" requires interpretation. Not parseable from a single source without reading prose.

P2-B Scope & Limits Definition — 3/5
Rate limits are partially stated: "1,000 requests/min" for standard operations. Pricing page lists execution quotas per tier (1,000 standard, 50 pro for Hobby). Overage pricing is explicit ($0.01/standard execution, $0.50/pro execution). Missing: per-tool rate limits, burst limits, API timeout specs. Scattered across pricing and docs rather than a single structured reference.

P2-C Substitution & Fallback Rules — 1/5
No documentation found for what happens when a tool is unavailable, a third-party API (GitHub, Slack, etc.) goes down, or execution quotas are hit. ToS disclaims uptime guarantees ("can't promise 100% uptime"). No fallback protocol stated.

P2-D Conditional Logic Transparency — 3/5
Pricing conditions are reasonably disclosed: Enterprise requires contact, startup program requires email request. Volume pricing is explicit. Conditions for audit logs and compliance are gated to Enterprise tier, disclosed on pricing page. Some conditions (specific OAuth limitations, tool-level permission scoping) require reading docs.

P2-E Semantic Precision — 5/5
Arcade uses precise technical language throughout: "authenticated tool-calling," "MCP runtime," "OAuth token management," "user-specific permissions vs. service accounts," "contextual access controls." Avoids vague superlative claims. Value props are technically specific and verifiable.

P3-A Verifiable Performance Data — 3/5
Public status page at `status.arcade.dev` showing 100% uptime across all components over 90 days, with per-component monitoring (Engine API, Dashboard, Cloud Worker, Cloud API, 19+ third-party integrations). Self-hosted, not third-party verified. No G2 reviews for arcade.dev specifically (0 reviews). Product Hunt listing exists with minimal reviews. Good infrastructure monitoring; limited social proof.

P3-B Scoped Permission Model — 4/5
Strong agent-scoped permission model: agents act with user-specific permissions rather than service accounts; token-level access controls; tool-level vs. server-level authorization documented; contextual access management. Enterprise tier adds RBAC and SSO/SAML. Missing: time-bounded or amount-bounded permission scoping for autonomous agents.

P3-C Audit Trail / Transaction Log — 3/5
Audit logs and compliance reporting are mentioned as an Enterprise-tier feature. Growth tier does not include audit logs. No machine-accessible audit API documented for agent consumption. Human-readable dashboard logs likely exist; agent-accessible log API not evidenced.

P3-D Behavioral Consistency Signals — 3/5
Weekly changelog published (Thursday cadence), spanning June 2025–March 2026. Uses semantic versioning for SDKs (v1.x, v2.x). Engine 2.0.0 had a breaking change (OAuth flow change). No explicit backward-compatibility policy or deprecation notice period stated in ToS. Change cadence is strong; formal stability guarantees are absent.

P4-A Friction-Free Activation — 4/5
"Try it free" sign-up flow available; no sales call required for Hobby or Growth tiers. API key available post-signup. Enterprise requires contact. The Hobby tier has a startup program (email required). Not fully instant (email verification likely), but self-serve for the primary tiers. Deducting 1 for the startup program email gate.

P4-B Agent Decision Signals — 4/5
Free tier with explicit execution limits (1,000 standard, 50 pro) provides clear agent-legible trial signals. Growth tier overage pricing ($0.01/$0.50) is machine-parseable. Pricing page is explicit about when to upgrade. Missing: a programmatic "check my usage" API signal that an agent could query to decide whether to upgrade automatically.

P5-A Integration Depth / Switching Cost — 3/5
7,000+ pre-built tools and integrations create meaningful switching cost. Custom MCP server builds using Arcade's SDK create codebase dependency. OAuth token storage in Arcade's system creates data gravity. However, MCP is an open standard — a competitor offering MCP runtime could theoretically substitute.

P5-B Agent Memory / Personalization Layer — 2/5
No dedicated agent memory or personalization layer evidenced. Tool execution history likely exists in logs (Enterprise tier), but no agent-accessible memory API documented. Each agent interaction appears stateless at the platform level.

P5-C Programmatic Renewal Signals — 2/5
Standard billing auto-renewal for Growth ($25/mo) is implied but not explicitly documented as agent-accessible. No renewal API or machine-readable billing status endpoint found.

P5-D Compounding Value Signal — 3/5
The pre-built tool catalog is described as growing (7,000+ integrations), creating compounding value for existing users. New tools automatically become available. However, no agent-readable signal communicates "new tools available since you last checked" or capability expansion metrics.

Rubric v1 (April 2026). Scores reflect the company's state on the audit date and may have improved since.