concepts

Pluckor vs. Claude for Chrome

Claude for Chrome puts Claude inside your browser; Pluckor hands your agent its own separate, scraping-grade browser as an MCP tool. Complementary, not competitors.

Short version: Claude for Chrome puts Claude inside your browser. Pluckor hands your agent its own separate browser as a tool. One is an assistant experience in the browser you already use; the other is developer infrastructure — an MCP tool surface any agent programs against. They’re complementary, not competitors.

The differences that matter

Claude for ChromePluckor
Whose browserYour everyday Chrome — your real profile, tabs, loginsA separate, dedicated Chrome for Testing it launches and owns (its own persistent profile)
What it isA Claude product feature — you pairing with Claude in your sessionAn open-source MCP server (npm) — a building block you wire into an agent
Who drives itClaude, specificallyAny MCP agent/model — Claude Code, Cursor, Codex, your own harness
Primary jobGeneral browsing & task assistance on the sites you’re usingData extraction / automation — especially getting through anti-bot
Anti-bot postureOperates your real session; not built to evade site defensesThe whole moat: reads run via content-script/tab APIs with no CDP, no automation fingerprint — engineered to get past Cloudflare and headless/Playwright blockers
Isolation & scaleYour one browserEach plk mcp process is its own lane with its own default tab, open_tab gives any agent a private handle, and PLUCKOR_INSTANCE gives it a fully separate browser — fleets of agents on one host
RepeatabilityEach task is a fresh conversationA known workflow becomes a script: replayed in one call, guarded by assertions, resumable from the exact step that broke
Where it runs / who owns itAnthropic-hosted product, in your daily browserSelf-hosted, 127.0.0.1-only — you own the browser and the data

When to reach for which

  • Claude for Chrome — “Help me in my browser”: summarize what’s on screen, fill this form in my logged-in session, drive a task across my open tabs. Human-in-the-loop, your real accounts.
  • Pluckor — “Give my agent a browser as a tool”: scrape a Cloudflare-gated catalog, extract structured data at scale, run many agents against many sites in isolated sessions — programmatic, disposable, repeatable.

Two pieces of Pluckor’s design carry most of this distinction. The reads-vs-interactions split is why reads leave essentially no automation fingerprint — no CDP, no navigator.webdriver — which is what gets Pluckor past Cloudflare and headless blockers. And separate instances plus per-connection lanes and explicit tab handles are what let a fleet of agents run side by side on one host. (Lanes are per plk mcp process — agents that share a connection, as subagents usually do, must each open_tab and drive their own handle.)

The takeaway

They solve different problems. Claude for Chrome is about you + Claude in your session. Pluckor is about handing an autonomous agent a clean, scraping-grade browser it can drive over MCP — with the fingerprint discipline, structured-extraction tools, and multi-agent isolation a data/automation workload needs and a personal-assistant experience doesn’t.

The soundbite: “Claude for Chrome automates your browsing; Pluckor is a browser your agent can automate.”