Compare

MCP-native slots instead of ephemeral browser-hours.

Shade is not trying to be a general REST API first. It is a remote MCP browser fleet where one connection gives agents named identity slots, saved sessions, structured supervision events, and human takeover.

Comparison reviewed June 2026. Competitor offerings change; check their current docs before buying.

CapabilityShadeBrowserbaseSteelPlaywright-MCP
Billing mental modelPersistent identity slot: saved sessions, consistent fingerprint, logins persist across runsUsage/browser-hour pricing with session and context primitivesUsage/browser-hour pricing with session and profile primitivesLocal browser process and host resources
Primary integrationMCP-native remote fleet over one connectionBrowser platform with REST/SDK APIs and a Stagehand-based MCP integrationSessions API and SDK-first browser cloudLocal MCP server launched per client
Identity continuityNamed slots designed to survive agent runsContexts can reuse cookies, authentication, and application data across sessionsProfiles and auth-context reuse can carry browser state across sessionsDepends on local profiles and client setup
Fleet visibilityOne console for up to 20 slots plus human takeoverSession live view, recordings, replays, logs, and inspector toolingLive and past session embeds plus agent trace surfacesLocal MCP introspection over the browser it launches
Token profileSynthetic action view from structured events; benchmark coming soonBrowser/session telemetry depends on the chosen Browserbase integrationSession views and traces depend on the chosen Steel integrationAccessibility-tree snapshots are the core MCP interaction surface
Abuse-sensitive featuresOpt-in personality, leak checks, no CAPTCHA solving, no residential proxiesAgent identity, verified browser, proxy, and CAPTCHA capabilities documented by BrowserbaseStealth, proxy, and CAPTCHA capabilities documented by SteelNo hosted anti-detect service

Token-efficiency wedge

Watch the fleet without paying snapshot tax

Accessibility snapshots are useful when an agent needs page structure, but they are a heavy supervision primitive when a human only needs to watch progress across many slots. Shade's console reconstructs activity from structured events: navigation, clicks, form edits, downloads, wallet prompts, leak checks, and takeover state. That gives humans a synthetic action view without feeding every page tree back through the model.

Measured June 11, 2026 across 7 real pages (Wikipedia, GitHub, Stripe, NYTimes, MDN, HN, python.org): Shade's browser_snapshot is hard-capped at ~1,500 tokens regardless of page weight - 3-13x smaller than an uncapped accessibility tree (the same Playwright API playwright-mcp emits) and 100x+ smaller than raw HTML. On nytimes.com that is 1,281 tokens vs 553,782 for raw HTML. Reproducible: benchmarks/token_efficiency.py.