PlugWiki
DesktopCommanderMCP logo

Desktop Commander MCP Server: Setup for Claude, Cursor and Other Apps

CommunityLocal
Getnpm@wonderwhy-er/desktop-commanderwonderwhy-er/desktopcommandermcp

Lets Claude and other assistants run terminal commands, read and edit files, and search your computer.

Desktop Commander gives your assistant the same reach you have in a terminal. It can run commands, keep programs running and talk to them, read and write files including Excel, PDF and Word, and search by name or content.

It runs locally and needs no account. It is a community project, not from Anthropic, and its own security page calls its limits advisory. Anything the assistant can do as you, this can do, so run it in Docker or a virtual machine if the assistant should not reach the rest of your computer.

Good for

  • Developers who want Claude to run builds, tests and scripts on their own machine
  • Analysts who want help with CSV, JSON and Excel files without leaving the chat
  • Power users who can isolate it in Docker or a virtual machine

Works in

12 should work

Tap an app to see how to set it up and any known problems.

Install DesktopCommanderMCP MCP

Claude Code

  1. 1

    Copy this prompt.

    Add the DesktopCommanderMCP MCP server to Claude Code. Run this command:
    
    claude mcp add desktopcommandermcp -- npx -y @wonderwhy-er/desktop-commander@latest
    
    Then check that it loaded and tell me if I need to sign in or restart.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

The npx route runs with your user permissions. The Docker route follows the README: the container sees only folders you mount, so add -v /your/folder:/mnt/folder arguments for the folders you want it to use. Without mounts it cannot see your files.

Built from the package data we hold. Not yet tested by us.

Tools

  • Starts a program or terminal command and waits for it to be ready

    Runs a command or program on your computer. It notices when the program is ready for input, so the assistant can keep talking to it.

    Try: “Start a Python shell and check my pandas version.”

    Runs code · start_process

  • Sends input to a running program

    Types into a program that is already running, such as a database shell or an SSH session, and returns what it replies.

    Try: “Run SELECT count(*) FROM users in the open psql session.”

    Runs code · interact_with_process

  • Reads new output from a running program

    Fetches output from a running process, with offset and length controls so long logs do not flood the chat.

    Try: “Show the latest output from the dev server.”

    Runs code · read_process_output

  • Reads a file, PDF or spreadsheet, or a web address

    Reads text files, Excel files, PDFs and URLs, with line or page controls. A negative offset reads from the end, like tail.

    Try: “Read the last 100 lines of server.log.”

    Files · read_file

  • Reads several files at once

    Returns the contents of many files in one call.

    Try: “Read package.json and tsconfig.json together.”

    Files · read_multiple_files

  • Lists the files in a folder and its subfolders

    Returns a recursive listing, two levels deep by default, with protection against huge folders.

    Try: “What is in my Documents folder, two levels down?”

    Files · list_directory

  • Shows details about a file or folder

    Returns metadata such as size and dates, and sheet information for Excel files.

    Try: “How big is report.xlsx and what sheets does it have?”

    Files · get_file_info

  • Searches files by name or by what is inside them

    Starts a streaming search over file names or contents, including inside Excel files. Results come back page by page.

    Try: “Find every file under src that mentions API_KEY.”

    Files · start_search

  • Gets the next page of a running search

    Returns more results from a search that is still running, using an offset.

    Try: “Show the next 50 search results.”

    Read · get_more_search_results

  • Lists searches that are still running

    Shows every active search session.

    Try: “What searches are still running?”

    Read · list_searches

  • Stops a running search

    Ends an active search cleanly.

    Try: “Stop that search.”

    Runs code · stop_search

  • Creates or changes a file

    Writes a file, replacing it or adding to the end. It also writes Excel files from a 2D array. The default write limit is 50 lines per call.

    Try: “Create notes.md with today's meeting summary.”

    Write · write_file

Read from the source code

Permissions

high risk · 5 things to watch · 2 advisories

High risk

The assistant can run any terminal command, change or remove your files and stop programs, with no read-only mode. Run it in Docker with only the folders you choose mounted, and approve tool calls instead of auto-approving them.

What it can do

  • Read your data
  • Create and change things
  • Delete things
  • Read-only mode

Watch out for

  • It runs any command as you

    The assistant can start programs, edit files and kill processes with your permissions. Prompt injection from a web page or file could trigger that, so use Docker or a VM for risky work. github.com

  • The folder and command limits are not a sandbox

    Allowed folders and the command blocklist only reduce accidents. Terminal commands can reach other folders and get around the blocklist, as the project says itself. github.com

  • The assistant can change its own settings

    A tool sets the blocked commands and allowed folders. Approve each settings change before it runs. github.com

  • It sends anonymous usage data unless you opt out

    The project says the data is pseudonymous and holds no file contents, paths or command arguments. Turn it off by asking the assistant or setting telemetry to false. github.com

  • A local history file keeps what the assistant ran

    Commands and arguments are saved on your machine, so secrets typed into a command can end up in the history. github.com

Security advisories

Technical details
Talks totelemetry (opt-out)
Sign-in and keysNo credentials. The server runs with your user permissions.
Data retentionTool calls and their arguments are recorded in local history files on your machine.

Questions

Is Desktop Commander official from Anthropic?

No. It is a community project by wonderwhy-er under the MIT license. The README says it is for Claude, but it works with other apps that support local MCP servers.

Is it safe?

Only as safe as the setup. It can run any command as you. The project says its folder limits and command blocklist are guardrails, not a sandbox, and recommends Docker or a virtual machine for real isolation.

Can I run it in Docker?

Yes. The README gives a Docker setup with the image mcp/desktop-commander and an installer script that asks which folders to mount. Without mounts, the container cannot see your files.

Can I use it with ChatGPT or Claude on the web?

Yes, through its remote mode. You start a device command on your computer, sign in through your browser, then connect your AI at mcp.desktopcommander.app. Commands still run on your computer while that device command is running.

Does it send my data anywhere?

It sends limited, anonymous usage data unless you turn it off. The project says this excludes file contents, paths and command arguments. Local history of your tool calls stays on your machine.

Does it need Node.js?

Yes for the npx route, which needs a current Node.js. The Docker route does not.

Can it delete my files?

Yes. It has no file-delete tool, but it can run shell commands that delete files, and it can stop programs. There is no read-only mode.

Information

updated Nov 2, 2026

Publisher
Community
Package
@wonderwhy-er/desktop-commander
Registry
npm
Transport
stdio
Version
0.3.1
License
MIT
Last commit
Oct 30, 2026
Open issues
204
Listed in
Official MCP registry, Docker MCP Catalog

WebsiteGitHubnpmOfficial MCP RegistryChangelogIssues

Sources

Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.

Show all 3
  1. github.com · github.com
  2. github.com · github.com
  3. github.com · github.com

Facts from the official MCP registry, GitHub, npm, PyPI and the Docker catalog. Updated Nov 2, 2026.