Troubleshooting¶
Common issues + fixes. If you don't see your problem, run linkedin-mcp-health first and share the output.
Installation¶
pipx install fails with PEP 668¶
You're using an externally-managed Python (Debian 12+, Ubuntu 23.04+, macOS 14+).
Fix: Use pipx (already installed as a separate venv manager), or use a venv:
pip install fails with "No module named pip"¶
Python is too old (pre-3.4) or pip is missing.
Fix: Install Python 3.11+ via your OS package manager or pyenv.
linkedin-mcp-health: command not found¶
The CLI scripts weren't installed in your PATH.
Fix:
- pipx:
pipx ensurepaththen restart shell - pip:
python3 -m site --user-base→ addbin/to PATH - venv: source the venv's
activate
Web dashboard¶
Port 8080 already in use¶
Or find and kill the conflicting process:
Pages return 404¶
You may be accessing an old /static/X.html URL. The dashboard uses clean URLs (/jobs not /static/jobs.html). Check your bookmarks.
Pages load but look unstyled¶
The CSS failed to load. Check:
- Browser DevTools → Network tab → look for failed
.cssrequests - Console errors
- Tailwind CDN may be blocked — try a different network
Sidebar links don't work¶
You may have an old version. Update:
Browser login hangs at "Opening browser..."¶
Playwright browsers aren't installed.
Fix:
Or manually:
LinkedIn connection¶
"Invalid cookies" / "Session expired"¶
li_at cookie expired (90-day max). Re-import:
"Challenge required" / "Verify it's you"¶
LinkedIn detected unusual activity. Do not automate your way around this — let the user solve it manually:
- Open LinkedIn in your normal browser
- Complete the challenge
- Re-export cookies
Login works but API calls fail with 401¶
Cookie validation fails on the server side. Check:
- Are you using the right account?
--account company - Is the cookie encrypted with a different passphrase? Re-encrypt with current one.
- Has the cookie been corrupted in storage? Re-import.
Posting¶
"Safety gate blocked" on every post¶
Your caps are set too low. Check /safety page or:
Common caps: 5 posts/day, 30 comments/day, 100 likes/day. Adjust in /safety.
Post succeeds locally but doesn't appear on LinkedIn¶
LinkedIn API delay. Wait 60 seconds and refresh LinkedIn.
If still missing, check /audit for the actual response code.
Post contains links → account flagged¶
LinkedIn aggressively flags new accounts that post links. Mitigation:
- Wait 2 weeks after account creation before posting links
- Mix link and non-link posts
- Use a domain you've "warmed up" (posted from browser first)
Scheduling¶
Scheduled post didn't fire¶
Check the worker is running:
If not, start it:
Or restart the web server (worker starts with it).
"Queue is empty" but I scheduled something¶
The schedule is in a different account's queue. Use:
Schedule fired but post is in "gate_failed" status¶
The safety gate blocked it. View the reason in /audit or:
Jobs module¶
"No CV uploaded"¶
Upload at /jobs → Setup tab. PDF or DOCX only.
Match scores seem wrong¶
The keyword Jaccard scorer is fast but approximate. For more nuanced scoring, configure an LLM at /llm.
Apply fails with "easy_apply_only" mismatch¶
The job doesn't support LinkedIn's Easy Apply. Either:
- Set
easy_apply_only=Falsein jobs settings (will require manual finish in browser) - Skip this job
Cover letter looks generic¶
LLM is not configured. Add an LLM provider at /llm for personalized cover letters.
AI agent integration¶
"MCP server not found" in Claude Desktop¶
- Check
linkedin-mcp-install verify claude-desktop - Restart Claude Desktop (config changes need a restart)
- Check
~/Library/Application Support/Claude/claude_desktop_config.json(macOS) or equivalent path
Tools don't appear in Cursor¶
- Restart Cursor
- Settings → Models → MCP → confirm "linkedin-mcp-pro" is listed
- Check Cursor's log:
Cmd/Ctrl+Shift+P→ "Show Logs"
"Tool execution failed: not authenticated"¶
The agent doesn't have credentials. Either:
- Run the MCP server with auth: set
LINKEDIN_MCP_API_TOKEN - Or run on localhost (no auth needed)
Performance¶
Dashboard is slow to load¶
- Check disk:
df -h ~/.linkedin-mcp/— should have >1GB free - Check DB size:
du -sh ~/.linkedin-mcp/profile/state.db— archive old audit logs - Restart the worker:
linkedin-mcp-schedule worker --restart
High CPU usage¶
Usually the deadman watchdog or feed listener. Check:
Stop unneeded listeners via the engagement page.
Out of memory¶
LinkedIn MCP Pro uses ~150MB baseline + ~50MB per browser. If you have many concurrent browser sessions, reduce with --max-browsers 2.
Logs & diagnostics¶
# Live logs
linkedin-mcp-web 2>&1 | tee /tmp/linkedin-mcp.log
# System health
linkedin-mcp-health
# Detailed diagnostics
linkedin-mcp-doctor
For bug reports, include:
linkedin-mcp-healthoutput- Last 50 lines of logs
- Steps to reproduce
- OS + Python version
Getting help¶
- GitHub Issues
- Discussions
- Safety docs — read before reporting "I'm getting blocked"