Works with the AI Agents you already use
Orphan MCP Hunter
Finds the MCP servers and node processes your AI coding tools left running after their session died, with evidence per row. Nothing is stopped until you say which.
$7
Orphan MCP Hunter
Example session with this skill installed
My Mac has been slow all week. Check if Claude Code or Cursor left any MCP servers or node processes running behind.
- Read your context and instructions
- Compiled the orphan mcp hunter
Scanned 412 processes with ps -axo pid=,ppid=,lstart=,command= at Sat Nov 15 00:22. Applied the three Stray rules. Two findings, one of them a judgement call.
| # | Finding | PID(s) | Started | Evidence | Suggested action |
|---|---|---|---|---|---|
| 1 | claude-mem worker (orphan runtime) | 14964 | Oct 29 02:01, 17d ago | PPID 1, bun .../claude-mem/10.6.0/scripts/worker-service.cjs, matches /.claude/ | stop |
| 2 | claude-mem ×2 (duplicate) | 63584 (keep 50619) | Nov 13 18:00, 29h before newest | same name, two parent sessions | stop 63584 if that session is closed |
| 3 | playwright-mcp ×2 (duplicate) | 63760 (keep 50715) | Nov 13 18:00, 29h before newest | same name, two parent sessions | stop 63760 if that session is closed |
Not flagged, and why
obsidian-mcp×2: both started Nov 13 21:37, two seconds apart, one Claude Desktop session with two vaults. Wrapper (npm exec) and leaf (node .../obsidian-mcp) counted as one instance each. Not duplicates.claude daemon rununder PPID 1: that is where it belongs.- 6 Claude Desktop renderer helpers whose args contain
claude-mcp: Electron--type=processes, excluded.
Totals: 1 orphan, 2 duplicate instances, ~214 MB resident.
Rows 2 and 3 come from a Claude Code session started Thursday evening. If that terminal is still open, keep them. If not, say "stop 14964, 63584, 63760" and I will send SIGTERM, wait 3 seconds, SIGKILL anything still alive, then re-scan and show you what is gone.
Nothing has been stopped yet.
Connects securely to your tools. The creator never sees your data.
About this skill
The problem
Claude Code, Cursor, Codex and Claude Desktop start MCP servers per session and often do not stop them. The session ends, launchd adopts the server, and it stays resident until you reboot. A real audit of one laptop on an ordinary afternoon: 3× claude-mem, 2× context7, 2× playwright-mcp, 4× obsidian-mcp. Eleven servers, zero living parents.
ps aux | grep finds them if you already suspect they exist. This skill is for when you do not.
What it does
Your agent runs one ps command and applies the same three rules as the Stray macOS app:
- Orphan MCP server (strong). Parent PID is 1 and the command line matches MCP tooling (
/_npx/,mcp-server,-mcp,/.claude/,modelcontextprotocol). - Duplicate instances (warning). Same server name running under two different parents, started more than an hour apart. Two windows of one session are not duplicates.
- Orphan runtime (warning). A
node,bunordenoprocess reparented to launchd and up for more than 24 hours. A fresh dev server you are still using is not flagged.
Every row shows the evidence: PID, parent, matched pattern, start time, and the process named after what it runs (claude-mem, playwright-mcp), never a bare node.
What it does not do
It does not kill anything on its own. The report comes first, then the agent waits for you to name PIDs. Termination is SIGTERM, a 3-second wait, then SIGKILL only if still alive, and the PID is re-checked against the report before anything is sent. Editor helper processes are excluded.
What you need
macOS. A shell. Nothing installed: ps and kill are built in. No network access.
Also available
dev-port-finder (what is on :3000), dev-cache-reclaimer (free disk safely), launchd-orphan-cleaner, or all four as the stray bundle.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
Strictly for macOS. Requires Bash access. Does not flag orphan runtimes under 24 hours old to avoid killing active dev servers.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 2 days ago
Mac only
- Software
Node.jsContext7 MCPPlaywright
Python