computer-use-mcp — Watchlist

MCP100 Assessment

Watchlist / Tracking

Why we’re tracking it

Monitor this implementation for changes in adoption, MCP interface, maintenance and permission boundary.

Reviewed Oct 10, 2026 · Documentation review; complete setup workflow not independently tested

Best fit

Give MCP agents broad control over a computer desktop

What we know

💻 Give AI models complete control of your computer (probably a bad idea) Give MCP agents broad control over a computer desktop.

Maintenance / activity evidence

GitHub REST pushed_at 2026-09-09T22:20:12Z

Tracking evidence observed Oct 10, 2026. These dated signals do not establish current maintenance. Reviewed enrichment confidence: medium.

Verification gaps / watch-outs

Installation and comprehensive security have not been independently audited.
Scoped by connected account, local OS, and configuration; verify exact permissions before installation. Reported action classes: mouse/keyboard input; full desktop UI actions. Complete write authority is not verified.
Local desktop permissions and controlled/sandbox user account
Sensitive access: Entire desktop screen and input state

Author explicitly warns about model errors and prompt-injection risk; can execute damaging UI actions. Use sandboxed user account.

Full authority, access and evidence limits are available in Review context and Access & risk.

Popularity

383 repository stars

Stars observed Oct 9, 2026.

Repository total for domdomegg/computer-use-mcp.

All limitations / Review context

Has computer-use-mcp changed its specific MCP integration, install method or privilege scope, and does it justify shortlist promotion?

High-impact permissions explicitly documented; no known blocker to an accurately caveated public Radar listing in limited first-party review.

Author explicitly warns about model errors and prompt-injection risk; can execute damaging UI actions. Use sandboxed user account.
Scoped by connected account, local OS, and configuration; verify exact permissions before installation.
Local desktop permissions and controlled/sandbox user account
Entire desktop screen and input state
Reviewed write actions: mouse/keyboard input; full desktop UI actions
Sensitive access: Entire desktop screen and input state
Complete write authority is not verified.
Installation, client compatibility and comprehensive security have not been independently audited. Authentication and access depend on the configured services.
No exhaustive advisory or runtime review performed; never infer installation safety.
Write actions/tool inventory not confirmed against exact tool documentation; listed action classes remain illustrative or context-dependent. Empty inventory does not establish read-only.
Setup and runtime permissions are implementation- and account-specific; follow linked upstream documentation.
Repository pushed_at explicitly observed

Original activity observations

GitHub REST pushed_at 2026-09-09T22:20:12Z

Technical identity

github.com/domdomegg/computer-use-mcp

Exact implementation source: https://github.com/domdomegg/computer-use-mcp ↗
Reviewed resource type: MCP Server
Repository identity: domdomegg/computer-use-mcp

Evidence & Sources

Evidence basis: Recorded source evidence
Setup runtime verification: Not independently runtime-tested by MCP100 for this setup workflow.

Reviewed Oct 10, 2026 · Latest source observation Oct 10, 2026

GitHub repository ↗
domdomegg/computer-use-mcp
Recorded repository material and project history.

Setup · Watchlist

Install & Agents

Publisher instructions.

Source evidence observed Oct 12, 2026.

Additional setup details required — see the specific gaps below.

Warning

Powerful desktop-control capabilities; grant system permissions only after reviewing actual host prompts and use supervised test actions.
README does not establish mandatory OS-specific permission steps.
Client config JSON is composed from publisher-linked install-mcp generator metadata, not verbatim in README.
This command may start a stdio MCP server rather than registering it; show the publisher client config alongside it, not as a standalone complete integration step.
Use only the named client/schema. Replace illustrative credential/path placeholders with your own settings. JSON wrapper composed from the publisher-linked generator payload; not a verbatim README sample.
Exact vendor-specific credential revoke procedure not established from checked official Setup documentation; may be external or not applicable. Do not invent.
Replace the supplied credential/path placeholders privately; use only the named configuration schema.
MCP100 client-scope note: this Claude Code CLI example applies only to that client and the exact registered name in the command. Desktop, Cursor and other clients use their own configuration; do not run it for a differently named entry.

Install / connect
Publisher instructions
Publisher-specific setup path. Follow prerequisites, then register or connect to your MCP host.
publisher documented install or connect · Publisher instructions
npx -y computer-use-mcp
Configure / authenticate
MCP100 setup guide
Publisher recommends install-mcp host setup; registry package identifier `computer-use-mcp`. Requires OS screen/control permissions.
Advanced configuration · illustrative mcpServers json compatible clients only · MCP100 composed guide
illustrative mcpServers json compatible clients only · MCP100 composed guide
{ "mcpServers": { "computer-use": { "args": [ "-y", "computer-use-mcp" ], "command": "npx" } } }
Check connection
Suggested check
Discover available computer-use tools but DO NOT click, type or manipulate the computer as a smoke test.
Remove / uninstall / revoke
MCP100 setup guide
Remove host MCP registration and any granted OS permissions separately if no longer needed.
MCP100 client-scope guidance: each CLI removal example targets only its named client and exact registered entry. Package names, Desktop/Cursor configuration keys and Codex/Claude registration names may differ; use the entry you actually registered. Client removal does not revoke provider credentials.
claude mcp remove computer-use

Official Documentation & Sources

What it does

💻 Give AI models complete control of your computer (probably a bad idea) Give MCP agents broad control over a computer desktop.

Capability: Developer Tools
Resource type: MCP Server
Best fit: Give MCP agents broad control over a computer desktop

Representative uses

Inspect the desktop through computer-use tools
Automate keyboard and mouse actions
Run supervised workflows in an isolated account
Give MCP agents broad control over a computer desktop

Limitations

Author explicitly warns about model errors and prompt-injection risk; can execute damaging UI actions. Use sandboxed user account.
Setup and runtime permissions are implementation- and account-specific; follow linked upstream documentation.
Repository pushed_at explicitly observed

Reviewed access

Access & risk

Review the exact access boundary before connecting this MCP to accounts, credentials, or sensitive data.

Authentication: Local desktop permissions and controlled/sandbox user account
Permissions: Scoped by connected account, local OS, and configuration; verify exact permissions before installation.
Data handling: Entire desktop screen and input state
Cost: may incur external API or infrastructure charges; unverified
Risk and cautions: Author explicitly warns about model errors and prompt-injection risk; can execute damaging UI actions. Use sandboxed user account.

Access scope and cautions

Write actions: Not verified mouse/keyboard input full desktop UI actions
Sensitive access: Entire desktop screen and input state
Risk level: critical
Security notes: Author explicitly warns about model errors and prompt-injection risk; can execute damaging UI actions. Use sandboxed user account.

The reviewed evidence does not establish the complete write authority of this MCP. Consult the unresolved evidence before granting write access.

FAQ

Based on reviewed evidence as of 2026-10-09. Watchlist tracking is not a recommendation or a security certification.

Which tasks are documented for this implementation?

The reviewed tasks include: Inspect the desktop through computer-use tools; Automate keyboard and mouse actions; Run supervised workflows in an isolated account; Give MCP agents broad control over a computer desktop.

What setup and authentication are documented?

A complete connection procedure is not verified. Client notes: Command provided only where first-party command or package existence is established; verify the client configuration against upstream documentation. The author explicitly recommends supervision and sandboxed OS account due to prompt injection and model error risks. These facts do not supply missing installation commands, credentials or a verified authentication procedure. Authentication: Local desktop permissions and controlled/sandbox user account

Can it modify data or perform write operations?

Not verified. The available evidence does not establish the complete write authority. The partial inventory lists: mouse/keyboard input; full desktop UI actions. Recorded permission context: Scoped by connected account, local OS, and configuration; verify exact permissions before installation.

What data-handling boundary should I consider?

Entire desktop screen and input state Recorded sensitive-access context: Entire desktop screen and input state.

What limitations and risks should I check before use?

Author explicitly warns about model errors and prompt-injection risk; can execute damaging UI actions. Use sandboxed user account. Setup and runtime permissions are implementation- and account-specific; follow linked upstream documentation. Repository pushed_at explicitly observed. No exhaustive advisory or runtime review performed; never infer installation safety.