Let your coding agent see what’s running on your Mac

What an agent can’t see on its own

A coding agent sees your files and whatever its shell commands print. It doesn’t know that a next dev from yesterday is still on port 3000, that memory pressure is high because Docker is using 12 GB, or that its own last session left three watchers running. So it starts a second server on 3001, blames the code for a slow test run, or tells you everything is clean.

The Model Context Protocol fixes this by letting an agent call tools on a local server. A system-info server answers questions about the machine rather than the codebase.

The DIY route: allow read-only shell commands

The zero-install option is to let the agent run the commands you’d run. In Claude Code, allow them in .claude/settings.json so it doesn’t ask every time:

{
  "permissions": {
    "allow": [
      "Bash(lsof -nP -iTCP -sTCP:LISTEN)",
      "Bash(ps -axo pid,ppid,%cpu,rss,etime,command)",
      "Bash(vm_stat)",
      "Bash(pmset -g assertions)"
    ]
  }
}

This works, with limits. The output is raw, so the agent spends tokens parsing it and still only learns node and a PID, not which project. There’s no safe kill: once you allow kill, the agent can stop anything you own.

Using Stat Space as the MCP server

Stat Space includes an MCP server in its free tier. The statspace binary inside the app runs it over stdio with --mcp, and it reads from the running app over a local socket, so answers come from data the app already collects.

Claude Code

One line, available in every project:

claude mcp add --scope user statspace -- /Applications/StatSpace.app/Contents/Helpers/statspace --mcp

Check it with claude mcp list, or /mcp inside a session.

Codex

Add to ~/.codex/config.toml:

[mcp_servers.statspace]
command = "/Applications/StatSpace.app/Contents/Helpers/statspace"
args = ["--mcp"]

Cursor, Claude Desktop, Windsurf

Add to ~/.cursor/mcp.json (or claude_desktop_config.json for Claude Desktop):

{
  "mcpServers": {
    "statspace": {
      "command": "/Applications/StatSpace.app/Contents/Helpers/statspace",
      "args": ["--mcp"]
    }
  }
}

statspace mcp-config prints all three snippets with the right path for your install, and Settings › CLI & MCP has them ready to copy. Stat Space has to be running for the tools to answer.

The tools

Tool	What the agent gets
list_ports	Every listening port with process, framework, project folder, launcher, uptime, URL and whether it’s exposed to the network
whats_on_port	Who is on one port
kill_port	Stops the process on a port; without confirm: true it only reports what would be stopped
top_apps	Apps by CPU, memory or network, with helpers grouped under their app
system_load	CPU (total, per core, load average), memory and pressure, swap, network, disk, GPU
agent_sessions	Running and recently ended agent sessions with child processes, ports and leftovers
orphans	Dev processes re-parented to launchd, with memory, age and ports
cleanup_agent_leftovers	Stops what an ended session left behind; also needs confirm: true
keep_awake	Keeps the Mac awake for a duration, until a time or indefinitely
sleep_blockers	Apps holding assertions that stop the Mac or display sleeping
explain_slowness	Plain-English answer to “why is my Mac slow?” (Pro)
what_is_process	What a cryptic process name is

Prompts that use it well

Once it’s connected, you can ask in plain language:

• “Before you start the dev server, check whether port 3000 is free. If it’s ours from earlier, reuse it.”
• “The tests got slow. Check system load and top apps before changing any code.”
• “We’re done. List anything this session left running and ask me before stopping it.”
• “Keep the Mac awake for 2 hours while you run the migration.”

You can also put the first and third in your project’s CLAUDE.md or AGENTS.md so the agent does them by default.

From your own terminal

The same data is available to you with the statspace CLI: statspace ports, statspace load, statspace agents, statspace orphans, with --json on any command for scripts. More on the CLI and MCP feature.