Control a real Chrome browser to complete any task: fill forms, extract data, book flights.
Nothing to run from here
Ghost Hands is installed from its publisher's own source and answers where it runs, so this marketplace is not in the path of a single call. There is no address here to send one to, and a panel that pretended otherwise would be showing you an answer we made up. Install it and call it from your own client — the Installation tab has the entry for each one.
94/100
trust score
3
downloads a day
local
runs on your machine
128
downloads in 30d
What it does
Governed browser hands for AI agents: classified actions, human approvals, replayable audit trails.
Quickstart
# 1 — add it to Claude Code
claude mcp add --scope user ghost-hands-mcp '--' uvx ghost-hands mcp
# 2 — ask your agent for something these tools do
# hands_perceive, hands_act, hands_run and 1 more
Ghost Hands costs nothing on mcprush: there is no plan to choose here, no cap to set and nothing mcprush can bill you for.
Publishernot claimed
LSAll 2 servers
littlestjames82-sys2 servers
2 categories
Collected from a public index. Nobody has claimed this account, so nothing here was written by its author — claim it if it is yours.
Installation
Where are you running it?
Every route below installs the same thing and ends at the same approval screen. This one runs on your machine: your client starts Ghost Hands as a process under your own user, with your files and your network, so the tool surface below is what it can do to your computer rather than to a server somewhere else. Its scan result is on the Security tab — read the surface before you approve it.
Claude Codedeclared by this listing
This is a public server: you run it yourself and this marketplace is not in the path. You need uv (docs.astral.sh/uv): uvx comes with it. Claude Code registers it in one command; --scope user makes it available in every project. The '--' is in quotes so that PowerShell passes it on.
claude mcp add --scope user ghost-hands-mcp '--' uvx ghost-hands mcpReconnect, or start a new session, and the tools appear in the model’s tool list.
Free on mcprush, and it runs from its publisher’s own package: this marketplace is not in the path, so there is no key of ours to hold and no card to add.
What it asks of you
Nothing is required, but it can take GHOST_HANDS_DRIVER (Session body for the MCP server: fake (default), chromium, simphone, or android.), GHOST_HANDS_START_URL (Optional URL opened on the first perceive call when a real body is selected.), GHOST_HANDS_APPROVER (Set to phone to route approval asks through the PhoneApprover (requires the Android bridge env vars).) if you want to set them.
What it writes
One config entry your client uses to start the process locally. A local server runs with your file system and your network, which is why it is priced without metering.
What you approve
4 tools, with what each one reads, writes and reaches shown before you agree — the same list on every route above. Read the tool surface.
Tool surface
What the model actually sees. Descriptions are diffed on every release — see version history.
Read from its source code. A server that builds its tool list when it starts can list more tools than this.
Look at a page before touching it: open a URL on the session's browser body (or install a fake web via 'pages') and return the numbered element map, where every interactive element is listed with a number. Those numbers are the only valid targets for hands_act. Perceiving never clicks or types…
| Parameter | Type | Required | Notes |
|---|---|---|---|
| url | string | no | The page to open and map. Omit it when installing a fake web via 'pages' (the fake web's start page is used). |
| pages | object | no | A fake web to install instead of opening a real URL: an object mapping page names to page definitions (the demo/testing body). Mutually exclusive with browsing a live site. |
Perform ONE governed action (e.g. {"kind":"click","target":3}) on the most recently perceived page. Every action is classified before it runs (readonly / write / consequential) and recorded in the provenance trail before it executes. Consequential actions are denied without an approver (or decided…
| Parameter | Type | Required | Notes |
|---|---|---|---|
| action | object | yes | The action to perform, as an object with at least a 'kind' (for example click, type, navigate, scroll) plus its operands, such as 'target' (an element number from the last perceived map) or 'text' for |
Run a whole scripted sequence of steps through the governed Runner in one call: each step is perceived, classified, approved under the same governor as hands_act, executed, and written to the trail. Use this for a deterministic multi-step job (the shape a recorded trail exports as a replayable…
| Parameter | Type | Required | Notes |
|---|---|---|---|
| steps | array | yes | The steps to execute in order. Each step is an action object like hands_act takes, optionally preceded by perceive steps; the Runner stops according to the session's stop rules and reports why. |
| pages | object | no | A fake web to install for this run (same shape as hands_perceive's 'pages'). Used with driver 'fake'. |
| start_url | string | no | Page to open before the first step runs. |
| driver | string | no | Which body executes the run: 'fake' (built-in demo web), 'chromium' (real browser), 'simphone' or 'android' (phone bodies). Defaults to the session body from GHOST_HANDS_DRIVER. |
| goal | string | no | A short statement of what the run is trying to achieve. Recorded in the trail as the run's purpose. |
| approve_all | boolean | no | When true, approval-gated steps in this run are approved without asking the approver channel. Only set it when the operator has explicitly authorized the whole run. |
Return this session's provenance trail: every decision and action recorded before execution, in order, with per-run accounting (steps, actions by class, stop reason). Read-only and local to this server session. Use it to audit what the hands actually did, or as the source for exporting a…
No parameter schema on file.
Free
$0 forever
- Every tool, no call limit
- No card, and mcprush charges nothing for it
- Source published under MIT
- Runs on your machine — nothing of it reaches our gateway
- Nothing to cap, because mcprush bills nothing
What counts against your monthly calls
| Tool | Unit | Calls used | Out of the allowance |
|---|
Nothing here is billable: mcprush charges nothing to install Ghost Hands or to call it, at any volume.
Call allowance · this installenforced at the gateway
Monthly allowance, in calls
—
—
Alert at 80% · then the tool returns a normal MCP error
Used this month: —
ATrust score 94 / 100
Release 0.9.3 · scanned 5 Nov 2026
Two independent axes, because powerful and malicious are different questions. The grade is threat only. The capability level is blast radius, and it is never a penalty on the grade — it is priced as one subtract-only term in the score, where you can see it.
Threat score100
Client score94
Findings10 threat · 0 capability
Capabilityhigh
Coveragesource
Verificationsource
Surface digest692436e3b5b12a6c68ee15c5e0bf91f82af57e8514ddf60d056dcc0b0b10d7c5
Score vectorthreat 100 → client 94
| Term | Level | What it prices | Points |
|---|---|---|---|
| capability-exposure | high | capability blast radius (high) — client exposure if the model is manipulated | −6 |
| verification-discount | source | publisher verification (provenance) — cryptographic build provenance ties the artifact to its source | −0 |
| coverage-honesty | source | inspection depth (source) — how much of the target the scan could see | −0 |
score = clamp(round(threatScore − E_cap − E_ver − E_cov), 0, 100). No model, no opinion, no network call at scoring time: the same methodology version against the same bytes always produces this number, and every line above is a field of the report rather than a summary of it.
What the scan could actually read
A grade is only as meaningful as its coverage, so the scanner publishes its own depth before it publishes its result.
Tool surfacenot available
Implementation sourceread
Package metadataread
Live transportnot available
No tools were enumerated, so prompt-injection, capability and toxic-flow analysis had no tool surface to inspect. To grade a package’s real runtime tools, scan the running server: --command "npx -y <package>".
Capability — what it could do if the model were manipulated
Tags derived from each tool’s schema and the implementation, not from what the tool calls itself. high is the level these add up to.
| Tool | Capability tags | Why the tag was assigned |
|---|---|---|
| No per-tool rows for this release — see the note below. | ||
can execute shell commands or code
Toxic-flow graph
The lethal trifecta, checked as a graph rather than as a checklist: untrusted input, a sensitive source and an external sink have to meet before there is a path worth worrying about.
The public result for this release does not print the flow graph, so there is nothing to show here. That is not the same as "no paths were found": what the scan did read is above, under coverage.
Supply chain and provenance
Integrityfirst-seen
Surface digest692436e3b5b12a6c
Scanned2026-11-05
This is the first scan of this surface here, so there is nothing yet to compare it against.
Every result on this tab comes from one deterministic pass over the published package — offline, rule by rule, and auditable line by line above. Same methodology version, same bytes, same score.
Release history
Always latest — the install command names no version, so it installs the newest release when it runs; the current release on file is 0.9.3. Pin a release below and the command asks for exactly that one.
0.9.3
undated
latestNo release note was published with this version.
—
no reviews yet
Who can review
One review per account, from a signed-in account that does not publish this listing. Nothing else is asked — you can say what you think before you install it. Publishers can reply once per review, and a review can be edited or deleted by whoever wrote it.
Sign in to write one
Any signed-in account that does not publish Ghost Hands can review it, once — before installing it or after. A review from an account that has installed it is marked as one.
No reviews yet
Nobody has reviewed this listing. The rating on the card is the mean of the reviews written here and nothing else, so there is no rating until somebody writes the first — which takes a signed-in account that does not publish it, and nothing else.