All 20 Criteria
P1-A Structured Data — 1/5
No evidence of schema.org markup, JSON-LD, or Organization/Product/Offer schema found on homepage or pricing pages. Sitemap exists but no structured data layer.
P1-B Machine-Readable Pricing — 3/5
Pricing table clearly presents four tiers (Free $0, Developer $20/mo, Startup $99/mo, Scale custom) in HTML with specific caps and overage rates. Not tagged with schema.org/Offer but numeric structure is machine-parseable.
P1-C llms.txt / Agent Layer — 2/5
A file exists at docs.browserbase.com/llms.txt but it functions only as a documentation link index without substantive content. No proper agent-facing identity or usage guidance. Not present at root domain.
P1-D API / MCP Availability — 4/5
Full REST API and SDKs for Node.js and Python. Published MCP server (`npx @browserbasehq/mcp-server-browserbase`) with 8 documented tools. Integrations with LangChain, crewAI, OpenAI, Claude, Gemini. No publicly linked OpenAPI/Swagger spec found, preventing a score of 5.
P1-E Discoverability (GEO) — 3/5
Strong presence in AI tool roundups, named in blog posts and framework docs. Used by Perplexity, Vercel, Microsoft, DeepMind. SEO-quality content exists across docs and blog. Not yet structured for AI retrieval optimization beyond standard SEO.
P2-A Offer Completeness — 3/5
What (headless browser infrastructure), who (AI agents/developers), and pricing (tiers visible) are findable but not co-located in a single machine-readable source. The pricing page and product page are separate; no single document captures the full offer.
P2-B Scope & Limits — 4/5
Pricing page explicitly states: browser hours per plan, concurrent browser caps (3/25/100/250+), session duration (15 min to 6+ hrs), proxy GB allotments, RPS limits per API, data retention windows (7–30+ days), and overage rates ($0.10–$0.12/hr). Strong coverage.
P2-C Substitution Rules — 1/5
No guidance found on service unavailability, fallback behavior, or substitution options. Status page documents incidents only after they occur with no pre-stated contingency.
P2-D Conditional Logic Transparency — 2/5
Scale tier is opaque ("contact sales," custom pricing). Enterprise conditions undisclosed. Free and paid tiers are well-defined but a significant portion of the offer surface is hidden behind sales.
P2-E Semantic Precision — 3/5
Mix of precise claims (2 vCPUs per instance, 15-minute session limit on Free, $0.12/hr overage) and vague marketing language ("robust," "reliable," "powerful"). Core technical specs are quantified; positioning copy is not.
P3-A Verifiable Performance Data — 2/5
Public status page (status.browserbase.com) exists with 5 monitored components and documented incident history since August 2024. However, no uptime percentage is published, no formal SLA is stated, and an independent benchmark (AIMultiple) reported a 50% task success rate in single-task evaluations, which undercuts self-reported reliability.
P3-B Scoped Permissions — 2/5
Authentication requires BROWSERBASE_API_KEY and BROWSERBASE_PROJECT_ID, creating project-level scoping. No evidence of agent-specific time-bounded, amount-bounded, or action-bounded permission tiers. SOC-2 Type 1 and HIPAA compliance exist but permission granularity is limited.
P3-C Audit Trail — 2/5
Session recording, command logging, live view iFrame, and source capture are available — but these are developer debugging tools. No machine-accessible audit log API for agent systems was found, nor evidence of structured transaction logs consumable by an orchestrating agent.
P3-D Behavioral Consistency — 2/5
SOC-2 Type 1 certification exists. Status page tracks incidents. No version-controlled terms of service, no published notice periods for API changes, and no documented stability track record for the API surface itself beyond the incident log.
P4-A Friction-Free Activation — 4/5
Free tier requires no credit card. MCP server installed via a single `npx` command. API key issued self-serve. Developer tier activatable at $20/mo without human approval. Activation path is close to instant with no sales gate for the first three tiers.
P4-B Agent Decision Signals — 2/5
Free tier serves as a trial signal (1 browser hour, 3 concurrent). No programmatic signals explicitly designed for agent consumption (e.g., no capability-check endpoint, no machine-readable upgrade triggers, no structured "when to scale" guidance).
P5-A Integration Depth — 3/5
Supports Playwright, Puppeteer, Selenium, LangChain, crewAI; open-source Stagehand framework creates additional ecosystem lock-in. Teams that adopt Stagehand or build production pipelines around Browserbase sessions accumulate meaningful switching cost.
P5-B Agent Memory Layer — 2/5
Session persistence is supported ("sessions that persist across runs so agents stay logged in") and sessions can run up to 6+ hours on Scale. But there is no rich, structured agent-accessible memory layer — no cross-session context store, preference API, or agent-identity persistence beyond browser cookies.
P5-C Programmatic Renewal — 1/5
Standard SaaS billing via subscription. No evidence of a renewal API, usage-triggered upgrade hooks, or any mechanism for an agent to programmatically manage its own subscription lifecycle.
P5-D Compounding Value Signal — 1/5
Session recordings and usage history accumulate, but no agent-readable signal surfaces compounding value (e.g., "your agent has solved 1,200 CAPTCHAs with 97% success — here's your efficiency trend"). Value accrues operationally but is not legible to agents.
Rubric v1 (April 2026). Scores reflect the company's state on the audit date and may have improved since.