desktop-commander
Use Desktop Commander MCP (typically tools like `mcp__desktop-commander__*`) for local files, precise edits, search, and long-running processes on the machine. Use it when the task requires direct local execution, rather than IDE-only actions or remote platform operations.
How do I install this agent skill?
npx skills add https://github.com/ry-run/run-skills --skill desktop-commanderIs this agent skill safe to install?
- Gen Agent Trust Hubpass
This skill provides a set of tools for local machine operations, including terminal command execution and file management. While it includes significant safety guardrails and transparency about its capabilities, the combination of external data ingestion (URLs and local files) with high-privilege execution tools creates a surface for indirect prompt injection.
- Socketpass
No alerts
- Snykwarn
Risk: MEDIUM · 1 issue
- Runlayerwarn
2/4 files flagged
- ZeroLeakspass
Score: 93/100 · 2 sections analyzed
What does this agent skill do?
Desktop Commander
Quick start
Goal: use Desktop Commander MCP to turn “files / processes / search / edits” into verifiable tool calls (small, safe steps) instead of treating the machine as a black box.
Most common entry points:
- Read content:
mcp__desktop-commander__read_file(paging, negative offset tail, PDF/image/Excel/URL). - Small edits:
mcp__desktop-commander__edit_block(targeted text replace / Excel range update). - Large edits:
mcp__desktop-commander__write_filein chunks (respectfileWriteLineLimit). - Interactive work:
mcp__desktop-commander__start_process+mcp__desktop-commander__interact_with_process+mcp__desktop-commander__read_process_output.
Official notes + tool list: skills/desktop-commander/references/desktop-commander.md.
Workflow decision tree
- Do I need to find something?
- File names/paths:
mcp__desktop-commander__start_search(searchType="files") →mcp__desktop-commander__get_more_search_results - File contents:
mcp__desktop-commander__start_search(searchType="content") → paginate →mcp__desktop-commander__stop_searchwhen done
- Do I need to read or change content?
- Read:
mcp__desktop-commander__read_file(useoffset/length; useoffset=-Nfor tail) - Small change:
mcp__desktop-commander__edit_block(default replaces 1 occurrence; useexpected_replacementsfor multiple) - Large change:
mcp__desktop-commander__write_file(mode="rewrite"thenmode="append"chunked)
- Do I need to run commands / keep sessions?
- One-off commands:
mcp__desktop-commander__start_process(shell command) + read output - REPL / SSH / DB / dev server:
start_process→interact_with_process→read_process_output
- Is this a high-risk operation (config changes, killing processes, bulk edits/moves, any data loss)?
- Explain impact + rollback first; require explicit user confirmation before executing.
- Prefer making config changes in a separate chat (official guidance).
Recipes
Reading files
- Text/code:
read_filewith pagination; logs:offset=-200(tail-like). - Multiple files:
read_multiple_filesto reduce round trips. - URLs:
read_filewithisUrl: truefor web content/images.
Editing files
- Targeted replace:
edit_blockwith minimal unique context; for many occurrences setexpected_replacements. - Rewrites:
write_filein 25–30 line chunks (rewritethenappend). - Excel: read via
read_file; edit viaedit_blockwithrange+ 2D array. - PDFs: only via
write_pdf(do not usewrite_filefor PDFs).
Search
- Prefer
start_search+get_more_search_resultsfor repo exploration; stop searches you no longer need. - Use
literalSearch: truefor patterns with special characters (parentheses, brackets, dots, etc.).
Processes & interaction
- Data analysis: run
python3 -i, then useinteract_with_processfor pandas/numpy workflows. - Observing long jobs: call
read_process_outputperiodically; to stop usekill_process/force_terminate(high-risk). - Status:
list_sessions/list_processes.
Config & audit
get_config/set_config_value: use carefully; directory restrictions are not a security boundary for terminal commands.get_recent_tool_calls: recover context and debug “what happened”.get_usage_stats: usage/performance insight.
Guardrails (must follow)
- Prefer absolute paths; don’t assume OS-specific separators.
- For big changes: read first; keep edits small; chunk writes; keep rollback in mind.
- High-risk actions require explicit confirmation: config changes, killing sessions/processes, bulk file edits/moves, any destructive command.
- Security:
allowedDirectorieslimits filesystem tools, not terminal commands—don’t treat it as sandboxing.
References
- Official notes + tool list:
skills/desktop-commander/references/desktop-commander.md
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/ry-run/run-skills/desktop-commander">View desktop-commander on skillZs</a>