Browser Use
Browser agents and stealth cloud browsers for AI: run web tasks, fill forms, extract data.
제작자가 쓴 설명(영어 원문)
Stealth browser automation for AI agents — real Chrome, N parallel persistent logged-in sessions.
이 세 가지는 절대 하나의 점수로 합치지 않아요. 의견, 측정, 인기는 서로 다른 것이니까요.
이 서버는 아직 쉬운 말로 된 요약을 쓰지 못했어요. 아래는 레지스트리에서 읽어 온 사실이에요.
쓰고 있는 AI 앱을 고른 다음 설정을 복사하세요.
필요한 것
어떤 앱을 쓰세요?
위에서 앱을 고르세요. 기억해 두었다가 다음부터는 버튼에 “…에 설치”라고 표시해요.
Claude Desktop → Settings → Developer → Edit Config를 열고 claude_desktop_config.json에 이 내용을 붙여 넣으세요
{
"mcpServers": {
"vibatchium": {
"command": "uvx",
"args": [
"vibatchium",
"mcp"
]
}
}
}터미널에서 이 명령어를 실행하세요
claude mcp add vibatchium -- uvx vibatchium mcp프로젝트의 .cursor/mcp.json 또는 ~/.cursor/mcp.json에 붙여 넣으세요
{
"mcpServers": {
"vibatchium": {
"command": "uvx",
"args": [
"vibatchium",
"mcp"
]
}
}
}브라우저가 앱을 열지 물어본 다음, 앱이 설정을 보여 주고 확인을 요청해요. 키 값은 비워 두니 직접 입력하세요.
프로젝트의 .vscode/mcp.json에 붙여 넣으세요
{
"servers": {
"vibatchium": {
"type": "stdio",
"command": "uvx",
"args": [
"vibatchium",
"mcp"
]
}
}
}브라우저가 앱을 열지 물어본 다음, 앱이 설정을 보여 주고 확인을 요청해요. 키 값은 비워 두니 직접 입력하세요.
이 서버는 내 컴퓨터에서 실행되는데, ChatGPT는 인터넷에 있는 서버에만 연결할 수 있어요. 다른 앱을 고르세요.
선택한 앱은 이 기기의 쿠키에만 저장돼요. 계정은 필요 없어요.
이 설정은 제작자의 레지스트리 정보로 만들었어요. README와 다르면 README를 따르세요.
내 컴퓨터에서 실행돼요. 설치하면 AI 앱에 도구 목록이 나타나요.
누구나 볼 수 있는 정보를 점검했고, 코드는 읽지 않았어요. 해당하지 않는 항목은 개수에서 뺐어요. · 2026년 11월 6일 점검
레지스트리 이름이 GitHub 계정 trueoriginlabs 기준으로 확인되었고, 이 계정이 저장소를 소유하고 있어요
아직 GitHub에서 이 저장소를 읽어 오지 않았어요
아직 GitHub에서 이 저장소를 읽어 오지 않았어요
내 컴퓨터에서 실행되는 서버라 인터넷 엔드포인트가 없어요
내 컴퓨터에서 실행되며, 대신 로컬 설정을 사용해요
내 컴퓨터에서 실행돼요. 도구 목록은 인터넷 엔드포인트가 있는 서버에서만 읽을 수 있어요
레지스트리에서 이 카테고리의 활성 서버 전체와 비교했어요. 스타는 게시자가 저장소를 소유한 경우에만 세요.
해당하는 기본 점검 1개 중 1개를 통과했어요. 코드는 읽지 않았어요. 키나 토큰을 넣기 전에 서버가 어떤 권한을 요구하는지 확인하고, 꼭 필요한 권한만 있는 키를 쓰세요.
레지스트리에 필수 키가 나와 있지 않아요. 다만 처음 쓸 때 로그인을 요구하는 서버도 있어요.
내 컴퓨터에서 실행돼요. AI 앱이 필요할 때 실행하고, 데이터는 제작자의 서버를 거치지 않아요(서버 자체가 온라인 서비스를 호출하는 경우는 제외).
GitHub 저장소는 마지막으로 – 변경됐고, 최신 버전(0.21.0)은 오늘 레지스트리에 올라왔어요.
Claude Desktop, Claude Code, Cursor, VS Code처럼 MCP를 지원하는 앱에서 쓸 수 있어요. 내 컴퓨터에 Python (uv)도 있어야 해요.
같은 카테고리에는: Browser Use, Chrome DevTools MCP, MCP Server Browser
같은 카테고리에서 다른 제작자가 만든 서버예요. 스타 순으로 정렬했어요.
카드 보는 법:★ GitHub 스타i세 기둥: 리뷰 · 점검 · 사용 정도i업데이트 = 최신 버전을 레지스트리에 공개한 날i
Browser agents and stealth cloud browsers for AI: run web tasks, fill forms, extract data.
MCP server for Chrome DevTools
MCP server for browser use access
제작자의 저장소에서 쓰인 그대로 가져왔어요. PopNx MCP가 검토하거나 보증한 내용이 아니에요. HTML, 이미지, 스크립트는 제거했어요. 링크를 누르면 이 사이트를 벗어나요.
The browser layer for an always-on box. Many logged-in identities, running unattended, all at once — on real Chrome, with stealth on by default, and with a security model for letting an AI near the credentials.
One daemon, N persistent Chrome sessions, a vb CLI and an MCP server. Self-hosted, free, Apache-2.0.
pipx install 'vibatchium[all]'
patchright install chrome vb setup # register the MCP server + agent skill (Claude Code, Codex, Cursor) vb explore https://example.com
| You want… | Use |
|---|---|
| An agent driving your signed-in browser while you watch | Claude in Chrome, Chrome Auto Browse — first-party, free, better integrated |
| The stealthiest single browser you can buy | CloakBrowser — fingerprints patched in Chromium's C++ (paid for concurrency) |
| The leanest driver for one agent | bladebro — five tools, one Rust binary |
| Browser automation without anti-bot pressure | playwright-mcp, agent-browser — bigger, older, better tested |
| Many accounts on one box, running unattended, against sites that fight automation | vibatchium |
That last row is the whole wedge. Supervised agents hand control back to you at every login wall and CAPTCHA; a library leaves the fleet — sessions, budgets, crash recovery, credentials — for you to build. vibatchium is that fleet layer.
As of 2026-10-06, from each project's own README, source and release notes (the MCP-cost row is our own measurement¹). Rows where we lose are included on purpose.
| vibatchium | CloakBrowser | bladebro | agent-browser | playwright-mcp | |
|---|---|---|---|---|---|
| N persistent logged-in sessions at once | yes, free, one daemon | 1 free (latest build) · more on Pro | one Chrome per agent session | yes, one daemon per session | one per profile dir |
| Unattended ops: crash self-heal, leases, budgets, idle freeze, goals | all of them | — | partial (crash relaunch, idle timeout) | partial (idle timeout, restore) | idle timeout |
| Agent can *use* a secret but not *read* it (origin-bound fill, read-back refused, TOTP) | yes | — | — | vault, origin check | --secrets redaction |
| Prompt-injection scanning of page content | yes | — | — | boundary markers (opt-in), no scanning | — |
| Browserless HTTP on the session's own cookies, Chrome TLS fingerprint · keyless web search | yes | — | — | — | — |
| Browser | stock Chrome | own Chromium build | stock Chromium | Chrome for Testing default | stock Chrome |
| Stealth layer | CDP-level (Patchright) | engine-level (87 C++ patches) | CDP-level ("6-layer"), no Runtime.enable | none in core; via cloud providers | none |
| Published stealth score | not yet | 30/30 on its own detector list | 68/80 Stealth Bench V1 (self-run) | — | — |
| MCP startup cost in Claude Code¹ | ~2.7k tokens (89 tools, 5 always loaded) | n/a | ~3.4k tokens (5 tools) | not measured | not measured |
| Platforms | Linux (macOS: CI probe, unsupported) | Linux · macOS · Windows | Linux · macOS · Windows | all | all |
¹ bytes/4 of tools/list plus server instructions, measured over stdio. Clients without tool search load every tool: vb mcp --caps min is 12 tools, ~3.8k tokens. bladebro's own README says ~1.9k.
Where the others are genuinely better: CloakBrowser's engine-level patches go deeper than any CDP-level layer can — and vb start --browser-binary will drive a Chromium build you bring. bladebro has a stealth benchmark number and we don't; until we do, read our stealth claims as measured on scoreboards, below, not on production walls. Both run on more operating systems.
1. A fleet of logged-in identities on one box. Each session is its own Chrome and profile, with its own proxy, timezone and GPU node, and — opt-in with vb persona set — its own screen and window, because by default every headless session on one machine reports the same 800×600 screen. vb fleet-check measures how alike your sessions still look. They run in parallel on one daemon that keeps them alive: a crashed renderer is relaunched on the same profile, parked sessions have their renderers stopped and thawed on the next call, leases stop two agents clobbering one session, and goals give a task a step/spend/time budget that survives a daemon restart.
vb --session alice start && vb --session alice go https://x.com # log in once, stays logged in
vb --session bob start && vb --session bob go https://x.com vb --session alice act "open notifications" & vb --session bob extract & wait vb fleet-check --sessions alice,bob # how twinned are they?
2. Credentials an agent can use but never see. The vault fills a password or a live TOTP code into the page without the value reaching the model's context, a screenshot or a log — and only into the site it belongs to. Over MCP, a prompt-injected agent can't type your GitHub password into a lookalike page, can't read it back with value or eval while it sits in the field, and can't edit the vault or pull a code. An agent with a *shell* is the operator — the CLI has deliberate escape hatches — so give agents MCP, not Bash, if that boundary matters to you.
printf %s "$GH_PASSWORD" | vb secret set github.com password --stdin
printf %s "$GH_TOTP_SEED" | vb secret set github.com totp-seed --stdin vb --session work fill @e7 --use-secret github.com:password vb --session work fill @e9 --use-secret github.com:totp vb wait-email-code github.com # operator CLI: polls IMAP and prints the code; refused over MCP
3. Page text is treated as untrusted. Every session scans the page text it returns for instructions aimed at the agent and flags them by default; wrap fences them off and redact removes them. Caller-supplied file paths are confined too, so an injected agent can't upload ~/.ssh or write into .git/hooks.
4. A browserless lane on the same identity. vb fetch reuses a session's cookies and proxy with Chrome's TLS fingerprint and client hints — JSON endpoints behind your login at curl speed, and it tells you when it can't match the browser. vb search finds URLs with no API key and no per-session budget.
vb --session work fetch https://api.example.com/v1/me
vb search "site reliability postmortem" -n 20 --urls
Status: alpha, active development. 2,331 tests green in CI (Linux, Python 3.11–3.14). Coding agents: read AGENTS.md first — the one-call recipes and the traps worth skipping.
Detector scores quoted below (bot.sannysoft, CreepJS, Cloudflare cold-launch) are manual observations, not CI-asserted — no test in the suite gates on them, and they are only as current as the last hand-run. The generated block under Measured scores is the one to trust.
pipx install vibatchium # core: browse / extract / screenshot / N parallel sessions
pipx install 'vibatchium[all]' # + fetch/search lanes, credential vault, VLM read, REST shim
patchright install chrome vb setup # register MCP + an auto-discoverable skill (idempotent) vb install # report which optional lanes are available
On a uv venv (no pip), add an extra with uv pip install --python /bin/python curl_cffi. Bleeding edge from master: pipx install 'git+https://github.com/trueoriginlabs/vibatchium#egg=vibatchium[all]'.
vb setup wires up the agents on *this* machine. To pull just the agent skill into any agent skills.sh supports (Claude Code, Codex, Cursor, Gemini CLI, Copilot, …), or to install skill + MCP server as one Claude Code plugin:
npx skills add trueoriginlabs/vibatchium # skill only, any agent
/plugin marketplace add trueoriginlabs/vibatchium # Claude Code: skill + vb mcp /plugin install vibatchium@vibatchium
Both still need the vb CLI on PATH (pipx install 'vibatchium[all]') — the skill drives it and the plugin's MCP server *is* vb mcp. On Claude Code, pick the plugin or vb setup, not both, or you register the server twice. The skill file is generated from vibatchium/setup_cmd.py; edit there and run python scripts/sync_skill.py.
vb explore https://example.com # one call: text-first, auto-closes
vb research --target https://example.com \ # parallel fan-out, N intents --intent "pricing model" --intent "customers" --intent "tech stack"
A wave of "headless browser for AI agents" tools rebuild the browser from scratch (Rust + V8, no Blink/Skia) to hit tiny memory and sub-100ms page loads. The catch is structural: with no rendering engine, they can't produce a real device's fingerprint — they synthesize one. And synthetic fingerprints don't hold still.
vibatchium drives *real* Google Chrome, so its fingerprints are real — and, more to the point, stable. The single test that separates the two is fingerprint stability across navigations. Run the same canvas + WebGL probe on two pages in one session:
| vibatchium (real Chrome) | synthesized-fingerprint engines | |
|---|---|---|
| canvas hash, page A → page B | identical | reseeded per navigation |
WebGL readPixels | real, deterministic pixels | often Math.random() |
| WebGL renderer | a real ANGLE renderer¹ | stub / zeros |
¹ Chrome's own software renderer (SwiftShader) by default — still a coherent, deterministic Chrome value, not a stub. A hardware-GPU string (e.g. ANGLE (Intel …)) needs the opt-in --gpu flag.
A real device returns the same fingerprint every page load; a fingerprint keyed off Date.now() does not — and *that inconsistency* is exactly what lie-detection fingerprinters (CreepJS and friends) flag. Measured: vibatchium's canvas hash and WebGL readback are byte-identical across navigations, and CreepJS reported 0 % stealth-tampering (no synthetic-environment signatures) when last run by hand. That figure is not regression-tested — treat it as an observation, not a guarantee.
This is not a claim of invisibility. The moat is fingerprint *authenticity*, not hiding that a browser is automated — vibatchium still reads as headless on the headless-specific tells (see Honest limits), and real-GPU WebGL (--gpu) is opt-in. But real, consistent fingerprints pass the consistency tier that synthetic ones fail *by construction* — and that tier is what stands between you and a login wall.
vb session new work
vb --session work start vb --session work go https://github.com # log in by hand once vb session new banking vb --session banking start vb --session banking go https://bank.example.com vb --session work click @e3 & # truly parallel — vb --session banking fill @e5 hi & # separate Chromes, no cookie bleed wait vb session list
Active-session resolution: --session FLAG → $VIBATCHIUM_SESSION env → ~/.config/vibatchium/active-session → default. Cap via VIBATCHIUM_MAX_SESSIONS=8 (default 8).
On one s
앞부분만 보여 드려요. 나머지는 저장소에서 읽어 보세요