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 Chrome | Pluckor | |
|---|---|---|
| Whose browser | Your everyday Chrome — your real profile, tabs, logins | A separate, dedicated Chrome for Testing it launches and owns (its own persistent profile) |
| What it is | A Claude product feature — you pairing with Claude in your session | An open-source MCP server (npm) — a building block you wire into an agent |
| Who drives it | Claude, specifically | Any MCP agent/model — Claude Code, Cursor, Codex, your own harness |
| Primary job | General browsing & task assistance on the sites you’re using | Data extraction / automation — especially getting through anti-bot |
| Anti-bot posture | Operates your real session; not built to evade site defenses | The 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 & scale | Your one browser | Separate profiles, isolated tabs/lanes, and fully separate instances (PLUCKOR_INSTANCE) — fleets of agents, no collisions |
| Where it runs / who owns it | Anthropic-hosted product, in your daily browser | Self-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 isolated tabs are what let fleets of agents run side by side without colliding.
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.”