SafariPilot
Provides an MCP server that controls the user's existing Safari session on macOS, enabling agents to manage tabs, navigate, read page content, click and type, fill forms, upload files, take screenshots, execute JavaScript, inspect console messages and network requests, and run batched browser actions.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
SafariPilot
SafariPilot is a local MCP server for Safari on macOS. Your agent works in the Safari you're already signed in to, in a separate window, and you keep using your Mac: it never brings Safari to the front, never switches your tabs, and never takes your cursor or keyboard.
Features
Your signed-in Safari. Your agent works in the Safari you already use, with your logins and cookies.
No cursor takeover. Your cursor doesn't move, keystrokes go only to Safari, and Safari can stay behind the app you're working in. Synthetic vs Trusted Input explains the trade-off.
Real clicks. Plain clicks and text input go through macOS Accessibility, so pages get trusted input. Key presses and double-clicks have caveats (see below).
Fewer round trips. Each action returns what it changed on the page, so the agent doesn't have to read the page again.
Related MCP server: MCP Safari Server
Requirements
macOS 26 or later
Safari 27 (earlier versions are untested)
A user logged in to the Mac's desktop. The app and Safari need a GUI session.
Local only: SafariPilot controls Safari on the same Mac as the MCP client. It can't drive a remote or cloud browser.
Install
brew install --cask johnsusek/tap/safaripilotOr download the DMG from Releases and drag SafariPilot to Applications. Releases are built, signed and notarized in GitHub Actions; to check that a DMG came from this repo's workflow:
gh attestation verify SafariPilot-*.dmg --repo johnsusek/SafariPilotMCPOpen SafariPilot. It lives in the menu bar; choose Settings… there to see its window. Then:
In Safari, open Settings > Extensions and turn on SafariPilot.
In SafariPilot's General pane, grant Accessibility.
In the Clients pane, set up your client: Claude Desktop, Claude Code, ChatGPT and Codex, Cursor, VS Code or LM Studio. For any other client, use Other Clients.
Ask your agent to open a page in Safari and click a link. Approve the client when SafariPilot asks.
Updates
SafariPilot checks for updates and offers to install them. Check for Updates… in its menu checks now, and the General pane turns automatic checks off. Homebrew installs can also update with:
brew upgrade --cask safaripilotBuild from source
Needs Xcode, Node.js and XcodeGen.
scripts/install.shThis builds SafariPilot, installs it in /Applications, opens it and links the safaripilot CLI into /opt/homebrew/bin.
Tools
The tools mirror those of Claude in Chrome.
Tabs:
tabs_list,tab_create,tab_select,tab_close,navigateRead:
read_page,find,get_page_text,screenshot,waitAct:
click,form_input,key,drag,scroll,scroll_to,file_upload,handle_dialog,resize_windowInspect:
javascript,read_console_messages,read_network_requests,page_fetchBatch:
browser_batchruns several calls in one round trip.
Agent permissions: An agent acts as you on every site you're signed in to. It can read pages, submit forms and upload files there, as well as run JavaScript and read network requests. Review which clients you approve, and take care with sensitive accounts.
Security and trust
Approve every client. Each new client needs your approval. A client you always trust gets a credential in its MCP settings, so other software on your Mac can't pass itself off as that client.
Tab isolation. Each agent session gets its own Safari window and sees only the tabs it opened, plus any popups from those; SafariPilot closes them all when the session ends.
Secure local connection. The Safari extension, the app and the agent talk only over loopback, and every connection needs a secret token that changes each launch.
Secrets stay hidden. Page reads, URLs and network logs redact password fields, card numbers, tokens and cookies.
Redaction isn't data-loss prevention. Screenshots,javascript results, page_fetch and other tool responses can still contain sensitive information.
Synthetic vs Trusted Input
Pages can tell real user input ("trusted") from events created in JavaScript ("synthetic"), and some sites reject synthetic events. Nearly all of SafariPilot's input is trusted. The exceptions:
Always synthetic: color inputs,
drag, and double, triple, right and modifier clicks. SafariPilot can't send these as trusted input without taking your cursor or making Safari the active app.Synthetic by default:
keypresses (notform_inputtext entry) andform_inputon dates, times, selects and sliders.compat: truemakes them trusted but brings the pilot window in front of your other Safari windows, because Safari sends real key presses and native menus only to its frontmost window. Safari as a whole can still stay behind other apps.
Limits
No hover: the CSS
:hoverpseudo-class is never applied.keyrefuses Cmd shortcuts so an agent can't trigger app or system commands with them.
Troubleshooting
In SafariPilot's window, General > Status shows Accessibility, the Safari extension and its connection, and Activity lists each call with the error code of any that failed.
Error code | Fix |
| Approve the client in SafariPilot's dialog. |
| Grant Accessibility in SafariPilot's General pane. |
| Allow the site from the SafariPilot button in Safari's toolbar. |
| Turn on the extension in Safari Settings > Extensions, then open a new tab. |
| Open Safari. |
| Safari is running the extension from another copy of SafariPilot. Remove the other copy, or run |
| Open SafariPilot from |
Benchmark
Claude Code (Sonnet 5.5) drove both browsers with the same prompts and no other tools. Of the 22 tasks, 15 run on public sites such as YouTube, DoorDash, the Apple Store and a PDF on irs.gov; the other 7 use local test pages. Each task ran 3 to 5 times per browser, and all 176 runs passed.
SafariPilot finished in under half the time, 1,176 s against Chrome's 2,471 s, with 32% fewer input tokens and 46% fewer output tokens. scripts/compare runs the tasks and draws the charts.
Tool times are the mean of every call in those runs. Claude in Chrome does its clicks, typing, scrolling and screenshots through one computer tool, so each of those calls counts under the SafariPilot tool that does the same job.
Comparison
SafariPilot | safari-mcp | MCPSafari | |
Makes Safari the active app | Never | Sometimes, then switches back to your app | Needs you to, for real input |
Puts its window in front of your Safari windows | Only with | No, it works in your window | When selecting a tab and for real input |
Changes the tab you're looking at | Never | For real input and screenshots | For new tabs and screenshots |
Takes your cursor or keyboard | Never | During file upload | During real input |
Competitors
safari-mcp has no trusted input in its everyday click, type and key tools; pages that check for it need its separate native tools. Since its tabs open in your window, real input and screenshots switch the tab you're looking at, and its AppleScript can bring Safari to the front. Passwords and tokens on the page reach the agent unredacted.
MCPSafari can't send a trusted click. Its real input needs Safari in front, and its hover and drag move your pointer. Its network log has no request bodies, and like safari-mcp, it never asks you to approve a client.
Apple's Safari MCP can't use your signed-in Safari; it opens a separate window without your logins.
How it runs
The MCP client starts safaripilot mcp, a stdio MCP server inside the app. That server passes each tool call to the SafariPilot menu bar app, which controls Safari through the SafariPilot extension and macOS Accessibility APIs. The server opens the app if it isn't running.
More
Architecture: how it works inside, building and testing.
Security policy: reporting a vulnerability.
License: MIT.
Thanks
To iMCP, whose app UI and client approval flow inspired SafariPilot's.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Stealth web browser for agents: search, fetch, click, download and type in persistent MCP sessions.
Search, read, and act on the web. The Stophy MCP server for agents.
Browz hosted MCP (free public beta): let an AI app drive your own Browz browser profiles from anywhere. Connect https://mcp.gobrowz.com/mcp and sign in with your Browz account (OAuth) or an agent token. Each call is relayed to the Browz app on your computer, which must be online with Remote control on and have a hardware security chip (Mac Secure Enclave or TPM 2.0). 17 tools: profiles, tabs, navigate, snapshot, act, wait, screenshot. Cookies, downloads and stored sign-ins stay on your computer.
Web scraping, browser automation, screenshots and AI data extraction for AI agents via MCP.
Related MCP Servers
- -licenseAqualityNot gradedmaintenanceA server that provides AI assistants programmatic control of Safari browser on macOS for web automation, testing, and debugging.161-
- AlicenseAqualityDmaintenanceEnables programmatic control of Safari browser on macOS for web automation, testing, and debugging via the Model Context Protocol.167 npm4MIT
- AlicenseAqualityDmaintenanceLets your coding agent control and inspect a live Safari browser on macOS. Provides access to Safari DevTools for debugging, automation, and testing.4951 npm3MIT
- AlicenseAqualityDmaintenanceMCP server that drives Safari on a physically attached iPhone via Apple's safaridriver, enabling navigation, screenshots, DOM snapshots, console logs, network timings, and tap/type/scroll actions on the real device. No Xcode, Appium, or WebDriverAgent required.161MIT