glovebox-mcp (segentic-lab/glovebox-mcp) is an MCP server listed on the M8ven Trust Index. It scores 51 out of 100, grade D. It declares 18 tools. No publisher has claimed this listing.

Back to MCP Trust IndexM8ven
D
Caution
51/100

glovebox-mcp

A sandboxed computer-use MCP server that lets an AI agent drive a real browser and desktop apps, confined to a nested X11 window so it cannot touch your real screen or files.

→ segentic-lab/glovebox-mcp· listed on glama

Caution. Specific findings reduced this grade. They are listed on the page. Grades reflect the full trust pyramid: code, verification depth, and reputation. New projects cap at C until adoption is earned.

How we verified

⚡ Live Monitored: not connected

Verified is a snapshot. Live keeps it current, and builds your track record.

⚡ Connect GitHub → continuous verification on every pushwhy connect →

Who stands behind it

segentic-lab

Source: Glama

Is this your MCP?

Claim it to get a verified publisher badge, a free copy of our full audit findings, and direct contact for any high-priority issues we find. Or connect your repo for our deepest verification, Live Monitored: read-only, revoke anytime. What we access →

⚡ Connect GitHub for Live Monitored

Install from

Glamanot scored here

The grade above is for the source repository. Registries can serve a different version, so we mark the ones we were not able to read.

// key findings
⚠️
Tool descriptions don’t match what handlers do
2 tools describe read intent but their handlers mutate — type_text (line 431: subprocess.run(["xclip", "-selection", "clipboard"],); open_file (line 477: subprocess.Popen(args, env=appenv, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL))
✅
No credential exfiltration, no sensitive file access, no obfuscation
Static analysis found nothing flowing your secrets to unexpected places.
✅
Open source with a license and README
Anyone can audit the code, the license is declared, and the publisher documents what it does.
// tools this server exposes18 tools

These names and descriptions are the publisher's own, read from the source code. We print them as written. Our assessment is the findings above, not this list.

launch_app

Launch a GUI app in its OWN new Xephyr display/window; returns its instance id. `command` is any shell command (e.g. 'chromium', 'xterm', 'gimp'). For Chromium, the X11 flags + a per-instance profile are added automatically. Control it with instance=<id>.

list_instances

List all instances (running app windows) with their display + name.

close_instance

Close an instance (kills its app + Xephyr display). Refuses instance 1 (the main sandbox).

get_screen_size

Screen size of an instance + active vision backend.

screenshot

Screenshot of an instance's window.

parse_screen

Detect on-screen elements (id, type, label, interactive, pixel-center) for an instance (backend = GLOVEBOX_VISION). Saves a numbered image to /tmp/glovebox_annotated_<N>.png.

click

Click at absolute coords. `observe`='screenshot'|'parse' returns the resulting screen state in the SAME call (saves a round-trip); `settle_ms` waits for the UI to update first. Default 'none'.

drag

Drag (mouse-down → move → up) from (x1,y1) to (x2,y2) — for drawing shapes / selecting.

double_click

Double-click at absolute coords in an instance.

click_element

Click an element id from the most recent parse_screen. `observe`/`settle_ms`: see click().

type_text

Type at the focused field — unicode-safe. Pure ASCII is typed via xdotool; text with any non-ASCII char (č/š/ž …) is inserted via the clipboard + ctrl+v, because xdotool's synthetic unicode keystrokes get silently dropped by some toolkits (e.g. Inkscape's GTK canvas) even though they work in browser…

upload_file

Attach a local file to a page's <input type=file> via Chrome DevTools Protocol — use this for ALL browser uploads (logos, images, docs). The sandbox's native GTK file picker is invisible to us AND hangs the renderer, so never click an upload button expecting a dialog — call this instead. Works on Ch…

open_file

Open a local file in an app ON an existing instance's display (NOT a new instance). If `app` is given, runs `<app> <filepath>` there (e.g. app='gimp'); otherwise tries xdg-open. Relative paths resolve against the instance's staging folder files/<N>/ (see list_files). For BROWSER <input type=file> up…

list_files

The instance's staging folder (files/<N>/ under the install dir) and its contents. Drop files there (or reference any host path you can read) to open in apps via open_file() or an app's Open dialog; browser uploads still go through upload_file().

press_keys

Press a key/combo (xdotool): 'Return','ctrl+a','Tab','ctrl+t'. `observe`/`settle_ms`: see click().

move_mouse

Move the mouse in an instance.

scroll

Scroll: positive = up, negative = down. `observe`/`settle_ms`: see click().

wait_ms

Wait (e.g. for a page to load). Capped at 10000 ms.

// environment variables
To run this server yourself, you supply these values. They go in your own MCP client configuration and stay on your machine. The secret label means the value is sensitive, not that the server mishandles it.
configGLOVEBOX_HOST_DISPLAYThe host display for new windows is (default :0); XAUTHORITY is auto-discovered.
configGLOVEBOX_VISIONYour choice is written to .vision-mode (override per run with the env var).
// quality suggestions

Tool annotations

No tools have read-only/destructive annotations

Add readOnlyHint or destructiveHint annotations to every tool so hosts can warn users before invoking.

All four hints declared on every tool

18/18 tools missing one or more hints — launch_app (missing: readOnlyHint, destructiveHint, idempotentHint, openWorldHint); list_instances (missing: readOnlyHint, destructiveHint, idempotentHint, openWorldHint); close_instance (missing: readOnlyHint, destructiveHint, idempotentHint, openWorldHint), +15 more. OpenAI's directory rejects tools where any of the four hints are missing or non-boolean.

For every tool, set all four hints (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) to explicit true/false values that match the handler’s actual behaviour.

Descriptions match behaviour

2 tools describe read intent but their handlers mutate — type_text (line 431: subprocess.run(["xclip", "-selection", "clipboard"],); open_file (line 477: subprocess.Popen(args, env=appenv, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL))

Rename the tool, rewrite the description, or move the side-effect into a separate clearly-named tool.

Tool inputs are validated

17/18 tool handlers declare input schemas (94%)

Declare an inputSchema with zod/joi/yup on every tool definition.

Tool handlers catch errors

Only 3/18 tool handlers wrap calls in try/catch (17%)

Wrap each tool handler body in try/catch and return a structured error response.

Tests exist

No test files found

Add tests that exercise each declared tool.

Tool description accuracy

4 tools have description/behavior mismatches: launch_app: description implies read-only but handler writes/deletes/executes; close_instance: description implies read-only but handler writes/deletes/executes; type_text: description implies read-only but handler writes/deletes/executes

Update tool descriptions to accurately reflect all capabilities — especially write, delete, or execute operations.

Claim the listing to review these findings one by one and send us a correction where you disagree, straight to the team. Claiming also means we tell you when the grade moves, and reach you first if we find anything urgent.

// full audit trail
The findings above are the summary. The full trail, every check we ran, each deduction, the network hosts observed and the dependency advisories, goes to verified publishers, along with an alert whenever a new one lands. Verified publishers can also review each finding and dispute it in one click. Publisher corrections have sharpened several of our checks this month, because the maintainer knows the codebase better than any scanner.
// improvement guidance — verified publishers only
We have 7 concrete improvements we can share with the publisher of this MCP. Each comes with specific guidance to raise the trust score.
// embed badge in your README
[![M8ven Verified](https://m8ven.ai/badge/mcp/segentic-lab/glovebox-mcp?variant=verified)](https://m8ven.ai/mcp/segentic-lab/glovebox-mcp)
Shows verification status without the grade. Want the grade badge instead? Remove ?variant=verified from the URL.
commit: f61bbc1883d3a213adc8f0f40c7499ac16f90328
code hash: f1ce7765f3a6bcc770a0748dc2a9e67c5f44cde9c5975bac65a157e6ee012f51
view raw JSON →
Check MCPs from inside your assistant
Tool Check · MCP

Vetting this one by hand? Tool Check is an MCP that scores other MCPs. Add it once and ask Claude, ChatGPT, or any MCP client to grade a server, surface CVEs, check the publisher, and suggest safer alternatives — before you install.

https://m8ven.ai/api/mcp/tool-check
check_toolsearch_toolscompare_toolsrecommend_alternativescheck_publisherreport_concern
How to add it →Free · no account needed · works in any MCP client