Repository navigation
stealth browser mcp review
This page is the project reviewing itself: it lives on invisible_playwright_mcp's wiki and is written by invisible_playwright_mcp's maintainers, so it is the most conflicted review of invisible_playwright_mcp you will find. It exists anyway because most of what ranks for "invisible_playwright_mcp review" is written by nobody in particular about a repository they have not read, and because a first-party review can do one thing a third-party one cannot: state plainly what the project does not do, and be accountable for it. Every factual claim below is checkable against the repository and the published package, both read on 2026-09-03 with the default model re-read on 2026-09-08, and the honest move for a reader is to treat the praise with suspicion and the self-criticism as reliable.
invisible_playwright_mcp is an open-source AI web agent: you describe a task in plain
language and it drives a real browser until the task is done. The
repository sits at about 30,300 stars and 4,600 forks, has existed since
August 2024, and is MIT licensed. The Python package invisible-playwright-mcp is on PyPI
and needs Python 3.11 or newer. This review was written against what the
index and the repository carried on 3 September 2026: 0.3.0 published,
0.4.0 on the main branch. The package has shipped many times since, so
the PyPI page is the current answer
and this paragraph is not; the statements below that name a version say
which one they describe.
One factual line on where it came from, because the repository description still says it: the project was born as an AI web agent that applied to jobs in bulk, drew mainstream press coverage in 2024 that was in part critical, and has since been rebuilt as the general-purpose agent this wiki documents; that earlier use is not covered here.
There are two ways to run it, and they share one browser:
-
Inside an assistant you already use. A plugin for Claude Code
(
claude plugin install invisible-playwright-mcp@oss-invisible, once the repository is added as a marketplace), an extension for Gemini CLI, or a config entry in Claude Desktop or Cursor, and your assistant's model does the thinking. -
Standalone.
uvx invisible-playwright-mcp uiserves a local page with chat on the left and the live browser on the right. It takes an OpenRouter key, defaults toz-ai/glm-5.3-flash, and accepts--modelfor anything OpenRouter serves. Since 0.3.0 this is the only invisible_playwright_mcp entrypoint; headless one-shots run through the assistant path above.
The differentiating bet is the browser itself. Instead of driving a stock automation build over an automation protocol, invisible_playwright_mcp drives a Firefox patched at the source level so that what a page inspects - the things JavaScript can read, the way input arrives - presents as a normal desktop machine. The agent also refuses to set form fields from JavaScript even when that would be quicker, because a page can tell the difference between script writes and typed input.
The browser story is real engineering, not a flag. The fingerprint
work lives in patched C++, is seed-deterministic (--seed gives you the
same browser identity on every run), and is maintained as its own
published engine. This is the part of the project we would defend in any
room.
It is honest about inputs. Keys are handled carefully: the OpenRouter key is stripped from the environment the browser process starts with, by name and by value, and a test in the repository fails if that stops being true.
Operational features match real use. --proxy takes HTTP or SOCKS5
and the browser's timezone, locale and egress follow it; --profile-dir
persists logins between runs; --headed shows the window; on the
assistant path the same knobs ride the MCP registration as
STEALTHFOX_* variables; and for scheduled work the engine itself is a
Python library with Playwright's API. The pieces you need to run it
repeatedly are there, not left as exercises.
It is genuinely open. MIT license, the engine and its wrapper published as installable packages, and the agent code in this repository short enough to actually read before you trust it with a browser.
This list is the reason the page exists. None of it is roadmap coyness; it is what the product is.
- It does not solve captchas. Nothing in the codebase attempts it. When a challenge appears, the agent's options are the same as any script's: stop, or hand the session to you.
- It does not promise you will not be detected. A patched browser changes what a page can read from the browser; it does nothing about your IP's reputation, your pacing, or per-account limits, and this wiki maintains a whole section on exactly those failure modes. Distrust any tool in this category that promises otherwise.
-
No macOS. Windows x86_64 and Linux x86_64/arm64 only; the last
macOS engine build was
firefox-20, and support ended. A Mac user cannot run the standalone product today. -
There is no free mode. Since 0.4.0
uvx invisible-playwright-mcp uirefuses to start without an OpenRouter key: an agent is a model with a browser, and tokens cost money. Driving the browser by hand without a model is the invisible_playwright library's job, not this product's. -
The browser is a quarter-gigabyte separate download that the server
fetches on its own when it starts, unless you pre-fetch it
(
uvx invisible-playwright fetch); on a slow connection the first task is answered with the download's progress rather than a page. - The model is not included, and results track the model. A weak model drives the good browser badly; browser problem or model problem is our own documentation of that boundary.
Sites can restrict accounts. Many platforms prohibit automated access in their terms. If you log the agent into an account on a platform that does, that account can be challenged, limited or closed, and no browser engineering changes the contract you accepted. Read the terms of the sites you point it at; the README says the same thing in its "Using it responsibly" section, and we mean it.
An agent with your session is an agent with your session. Anything
the logged-in browser can do, a misread page or a badly phrased task can
do too. Use --profile-dir deliberately, keep high-value accounts out of
it, and do not submit anything a human has not read.
Prompt injection is a live category. A hostile page can try to talk the model into actions you did not ask for. Big-vendor agents carry classifiers for this; invisible_playwright_mcp's standalone loop places that trust in the model you chose, and caution in the tasks you give it.
The UI binds to localhost for a reason. --host beyond 127.0.0.1
exposes an interface with no authentication; the README warns about
this, and so does this review.
For the searcher asking "is invisible_playwright_mcp safe" in the download-sense: the
canonical repository is
github.com/oss-invisible/playwright_mcp and the
canonical package is
invisible_playwright_mcp on PyPI. A project with this
history has accumulated forks and mirrors on other sites, some carrying
old code under the old license and the old purpose; none of them are
maintained here, and anything this wiki says applies only to the
canonical pair above. The code you run is auditable before you run it,
which is the strongest safety statement an open project can truthfully
make - not "trust us", but "check".
invisible_playwright_mcp is MIT licensed today. The relicense landed on 2 September 2026, and the README states the boundary precisely: everything distributed before 2 September 2026 was released under AGPL-3.0 and stays under it. In practice: the current code can be used, modified and embedded under MIT's permissive terms; old snapshots and forks made from them remain AGPL, with its copyleft obligations. If you vendored invisible_playwright_mcp code before that date, your obligations follow the license it shipped under, not the current one.
Use invisible_playwright_mcp if the browser is your bottleneck: your tasks are real browsing on sites that inspect their visitors, you want open code on your own machine, and you accept the token bill and the no-macOS line. Its engine is the most serious open-source attempt we know of at making the browser itself unremarkable, and we say that as the people who would.
Look elsewhere if you need macOS, captcha handling, a hosted
zero-setup product, or the largest possible community -
browser-use has an order of magnitude more
adopters, and the alternatives hub
compares the field with the same disclosure this page opens with. And
whatever this page just told you, it was the project grading its own
exam: run uvx invisible-playwright-mcp ui against a real task of yours, which costs an
afternoon and answers the only question that matters.
Is invisible_playwright_mcp legit? The canonical repo is oss-invisible/playwright_mcp (about 30k
stars, MIT) and the package is invisible-playwright-mcp on PyPI. It is real, maintained,
and auditable; mirrors elsewhere are not ours.
Is invisible_playwright_mcp safe? The code is open for inspection before you run it, and the API key is provably isolated from the browser process. The operational risks - account restrictions, prompt injection, what you log it into - are yours to manage and are listed above.
Is invisible_playwright_mcp free? The software is MIT-licensed free software. Model tokens are the running cost, via your OpenRouter key or your assistant's subscription.
Does invisible_playwright_mcp work on macOS? No. Windows and Linux only; macOS engine
builds ended at firefox-20.
Does invisible_playwright_mcp solve captchas? No, and it does not claim to. A challenge stops the run or goes to a human.
Who maintains invisible_playwright_mcp? The developer behind the feder-cr account,
who also maintains the patched-Firefox engine and this wiki - which is
exactly the conflict declared in the first line.
See also: The best model for an MCP browser agent, and what it really costs for the token-cost side of the decision, and Running invisible_playwright_mcp with Claude Code for the MCP route.
- The invisible_playwright_mcp repository: README, LICENSE,
pyproject.tomlandtests/test_key_isolation.pyread in the working tree on 2026-09-03; stars, forks, license and creation date read via the GitHub API the same day;src/invisible_playwright_mcp/llm.py, for the default model id, re-read 2026-09-08. -
invisible_playwright_mcpon PyPI, version 0.3.0 metadata checked against the index 2026-09-04. - The relicense commit ("Relicense under MIT", dated 2026-09-02) in the repository history, and the README's license section stating the AGPL-3.0 boundary for earlier distributions, both read 2026-09-03.
- For comparative claims about other tools, the pages linked above carry their own dated sources; none are repeated here.
This is invisible_playwright_mcp's wiki reviewing invisible_playwright_mcp. You have every reason to discount the compliments, so we put the limitations in their own section and made every claim point at something you can check without trusting us.
- OpenAI Operator alternatives
- Open-source Operator-style agents
- Is OpenAI Operator still available?
- OpenAI Operator vs Claude computer use
- browser-use alternatives
- Choosing an AI browser agent
- Open-source AI browser agents
- Open-source computer-use agents
- What is an AI web agent?
- AI browser agents vs traditional scraping
- Cloud browser infrastructure for AI agents, explained
- Browserbase alternatives
- Firecrawl vs an AI browser agent
- Skyvern alternatives
- Stagehand vs browser-use
- Project Mariner is gone: what replaced it
- Manus alternatives
- Gemini computer use vs Claude computer use
- A stealth browser MCP, reviewed honestly by its own wiki
- AI browser vs AI browser agent: which one do you want?
- AI browser agent vs RPA: which one fits the job
- AI browser agent vs n8n, Zapier and Make
- Vercel agent-browser alternatives, compared honestly
- What is an agentic browser? Definition and the two kinds
- Open-source agentic browsers: the three layers, compared
- Choosing an MCP server for browser automation: four axes
- Stealth MCP servers compared: Camoufox, nodriver, Patchright
- Playwright MCP alternatives, and the three you don't need
- Autonomous browser agents: the four rungs of autonomy
- What is actually free in the AI browser agent stack
- browser-use on GitHub: what the repo actually gives you
- Playwright MCP vs Chrome DevTools MCP: different jobs
- How to choose among MCP servers: a map by category
- Which MCP servers are worth adding to Claude Code
- MCP on GitHub: finding servers and judging them fast
- MCP vs an API: the decision, and what the wrapper costs
- MCP alternatives: when the protocol is the wrong shape
- Why does my AI agent get blocked?
- The timing signal AI agents give off
- Agent retry loops trip rate limits, not fingerprints
- Claude computer use detected as a bot
- browser-use getting blocked: what you can and cannot change
- Playwright MCP session blocked: four causes, four fixes
- Playwright MCP and captchas: what actually gets you past
- Cloudflare and a browser MCP server: what is being read
- Can an AI agent solve a captcha? The honest answer
- Getting an AI agent to fill out forms
- The best model for an MCP browser agent, and what it really costs
- Browser problem or model problem?
- How to use a Playwright MCP server with Claude Code
- Extracting data to a CSV with an AI agent
- Monitoring a page for changes with an AI agent
- How to let Claude Desktop control a browser
- How to add a browser to Cursor as an MCP server
- Using an AI agent to hunt for apartments
- Getting website data into Google Sheets with an AI agent
- Using an AI agent to download invoices from portals
- AI agents for web research
- Using an AI agent to test your own website
- How to add a browser to Cline as an MCP server
- Posting to social media with an AI agent
- Posting to Facebook with an AI agent
- Posting to Instagram with an AI agent
- Posting to X with an AI agent
- Automating LinkedIn posts: read this first
- Appointment bots: what they are and what an agent can legitimately do
- Track prices across sites with an AI agent
- Build a lead list with an AI browser agent
- Run an AI browser agent on a schedule
- AI browser agent with a local LLM: what changes
- Should you log your AI agent into your accounts?
- How to write a task an AI browser agent can follow
- Move data between two web apps with an AI agent
- The MCP server
- How the tools are shaped, and why
- Playwright MCP vs the Playwright CLI: which fits when
- Playwright MCP: browser is already in use, and the fix
- Playwright MCP best practices: four decisions that matter
- Playwright MCP with a proxy, and the three leaks it leaves
- A browser MCP server in GitHub Copilot: setup and limits
- Using a browser MCP server for web scraping: the pattern
- Which LLM for browser automation: the four properties
- How to build a browser agent, and what to take instead
- Getting an AI agent to log into a website: three routes
- MCP tools, resources and prompts: who controls each
- How many MCP tools is too many? The context arithmetic
- How to build an MCP server: the decisions, not the scaffold
- Local or remote MCP server: what changes, and what does not
- Writing an MCP client in Python: the thirty-line version
- Self-hosted AI agent: what one actually costs to run
- How long an AI browser agent takes per step, measured
- Text, HTML, snapshot or screenshot: what the agent should read
- Giving an AI browser agent a stopping condition
- Keeping an AI browser agent out of destructive actions
- Why did the AI agent click the wrong thing
- When the page changes under the AI agent
- Running one AI agent task across a list of sites
- Seeing a page as it appears in another country
- Getting data out of a dashboard with no export button
- Two browsers in one session: main and support
- Finding the dead links on a site with an AI agent
- Filling a CRM record from a company's website
- One form submission per spreadsheet row, with an AI agent
- Dated screenshots of a page as evidence
- Checking order and delivery status with an AI agent
- Reading a PDF that opens inside the browser
- Summarising a long page or thread with an AI agent
- Collecting every image on a page with its caption
- Collecting event and course listings with an AI agent
- Cancelling a subscription with an AI agent
- What an AI agent can and cannot do inside an iframe
- Shadow DOM and an AI agent: you can click it, you cannot read it
- How to add a browser to Codex as an MCP server
- What a page snapshot costs, per control
- How to let Gemini CLI use a browser
- Native selects and the ones that only look like selects
- Clicking by selector or by coordinates
- How long the agent waits before it gives up
- What a second browser costs
- Uploading a file with an AI agent, and why this one cannot
- Watching the agent work, and when it is worth it
- When not to use an AI browser agent
- Agent or script: deciding once instead of every time
- Using the keyboard instead of the mouse
- Secrets in an agent task: where they end up
- What an agent run should log
- Deduplicating what an AI agent collects
- Normalising values across sites
- Validating an AI agent's output
- Reading a table with an AI agent
- Driving a site's own search and filters
- The task works headed and fails headless
- How to let the AdaL CLI use a browser