proxy-doctor

proxy-doctor

Diagnose proxy misconfigurations that break AI coding tools.

When your browser works fine but Cursor / VS Code / Windsurf AI features don't — proxy-doctor tells you exactly why and how to fix it.

The Problem

AI coding tools (Cursor, VS Code with Copilot, Windsurf) rely on long-lived streaming connections (SSE/HTTP2) that break when:

• Your system proxy points to a localhost port where nothing is listening
• A VPN/proxy app was closed but its settings linger in macOS system preferences
• Your editor inherited stale proxy environment variables from launchctl
• The proxy is running but buffers streaming responses, breaking AI completions

The result: "browser works, AI editor doesn't" — the most common and frustrating developer experience.

What It Checks

proxy-doctor inspects 5 layers of your macOS proxy configuration:

Layer	What	How
1. System Proxy	Web/HTTPS/SOCKS proxy across all network services	networksetup
2. Residual Values	Disabled proxies with stale localhost addresses	Parse disabled-but-set entries
3. Port Health	Whether referenced proxy ports are actually listening	socket.connect()
4. Editor Config	settings.json, argv.json, recent error logs	File read + pattern match
5. GUI Environment	http_proxy/https_proxy in GUI app context	launchctl getenv

Quick Start

CLI

# Install
pip install proxy-doctor

# Run diagnosis (JSON output — default, optimized for AI agents)
proxy-doctor check

# Run diagnosis (human-readable output)
proxy-doctor check --human

# Show recommended fixes
proxy-doctor fix

# Check a different editor
proxy-doctor check --editor vscode

As an MCP Tool (for AI agents)

proxy-doctor ships as an MCP server that AI agents can call directly:

# Install with MCP support
pip install proxy-doctor[mcp]

# Run MCP server
python -m proxy_doctor.mcp_server

Add to your MCP configuration (e.g., Cursor ~/.cursor/mcp.json):

{
  "mcpServers": {
    "proxy-doctor": {
      "command": "python3",
      "args": ["-m", "proxy_doctor.mcp_server"]
    }
  }
}

Note: "command" must point to the Python where proxy-doctor[mcp] is installed. If python3 doesn't work, use the full path (run which python3 or python3 -c "import sys; print(sys.executable)" to find it).

Your AI agent can then call:

• diagnose_proxy(editor="cursor") — full 5-layer diagnosis with evidence
• list_fixes(editor="cursor") — recommended fixes with executable commands
• supported_editors() — list available editors

For AI Agents: Quick Install

If you are an AI agent setting up proxy-doctor for your user:

# 1. Install
pip install proxy-doctor[mcp]

# 2. Find the Python path
python3 -c "import sys; print(sys.executable)"

# 3. Add to ~/.cursor/mcp.json (use the path from step 2)
# {
#   "mcpServers": {
#     "proxy-doctor": {
#       "command": "<python-path-from-step-2>",
#       "args": ["-m", "proxy_doctor.mcp_server"]
#     }
#   }
# }

# 4. If MCP is not available, use CLI directly (zero dependencies):
proxy-doctor check          # JSON output
proxy-doctor check --human  # human-readable
proxy-doctor fix            # show fixes (read-only)
proxy-doctor fix --apply    # apply fixes (asks for confirmation)

Daemon Mode (v0.2+)

Run proxy-doctor as a persistent background service with automatic health monitoring:

# Start daemon (installs as macOS launchd service)
proxy-doctor daemon start

# Check daemon status
proxy-doctor daemon status

# Stop daemon
proxy-doctor daemon stop

# Check for updates
proxy-doctor update

The daemon runs every 5 minutes, compares results with the previous check, and sends a macOS notification when status changes (e.g. healthy → unhealthy).

Menu Bar (SwiftBar)

# If SwiftBar is installed
cp plugins/swiftbar/proxy-doctor.5m.sh ~/Library/Application\ Support/SwiftBar/Plugins/
chmod +x ~/Library/Application\ Support/SwiftBar/Plugins/proxy-doctor.5m.sh

Shows a green/red/orange indicator in your menu bar with one-click diagnosis.

Example Output

Unhealthy (Case A: dead proxy port)

{
  "status": "unhealthy",
  "diagnosis": {
    "case": "A",
    "root_cause": "Editor is configured to use proxy at 127.0.0.1:10903, but no process is listening on that port.",
    "confidence": "high",
    "source": "system proxy (Wi-Fi (http))",
    "browser_explanation": "Browser may use a different proxy path (e.g. browser-only mode) or fall back to a direct connection."
  },
  "fixes": [
    {
      "fix_id": "clear-system-http-wi-fi",
      "description": "Disable http proxy on Wi-Fi",
      "command": "networksetup -setwebproxystate \"Wi-Fi\" off",
      "risk": "low"
    }
  ]
}

Healthy

proxy-doctor v0.2.0
Editor: cursor | Platform: Darwin

Status: HEALTHY

No proxy contamination detected.

Supported Editors

Editor	Config Detection	Log Scanning	Status
Cursor	yes	yes	supported
VS Code	yes	yes	supported
Windsurf	yes	yes	supported
Claude Desktop	planned	—	future
Zed	planned	planned	future

How It Works

proxy-doctor identifies three failure patterns:

Case A — Dead proxy port (high confidence): Your system or editor points to 127.0.0.1:port but nothing is listening. This happens when a VPN/proxy app is closed but its settings remain.

Case B — Streaming broken (medium confidence): A proxy is running, but it buffers SSE/streaming connections that AI editors depend on. Common with browser-only proxy modes.

Case C — Path mismatch (medium confidence): Browser and editor use different proxy paths. Browser works via a dedicated proxy route; editor inherits a stale or incompatible one.

Platform Support

• macOS: Full support (system proxy, launchctl, networksetup)
• Linux: Partial (editor config + environment variables; no networksetup)
• Windows: Not yet supported

Trust & Permissions

proxy-doctor follows a read-only by default design. No system changes are made unless you explicitly opt in.

Default Behavior (read-only)

Access	What	Why
Reads	System proxy settings, editor config files, launchctl env vars, local port status	Core diagnostic functionality
Writes	~/.proxy-doctor/ only (cache, logs, update state)	Daemon state persistence
Network	pypi.org (version check only)	Auto-update feature
Does NOT	Modify proxy settings, change editor config, send telemetry, access credentials	By design

Opt-in Fix Application

proxy-doctor fix shows the recommended commands but does not execute them.

To apply fixes, use proxy-doctor fix --apply:

• Each fix is shown with its command and risk level
• You are prompted individually for each fix ([y/N])
• Default is No — nothing runs unless you type y
• You can abort at any time with Ctrl+C

This two-step model lets AI agents safely call list_fixes() via MCP to see what needs fixing, then present the commands to users for approval.

Feedback

Found a bug? Have a suggestion? Diagnosis didn't help?

• GitHub Issues: https://github.com/Jiansen/proxy-doctor/issues/new
• For AI agents: The JSON output includes a feedback.report_issue URL — include the full JSON when filing issues.

For AI agents providing feedback programmatically (zero dependencies):

# Create a GitHub issue via CLI (requires gh)
proxy-doctor check | gh issue create --repo Jiansen/proxy-doctor \
  --title "Diagnosis report: [describe issue]" --body-file -

# Or simply: copy the JSON output into a new issue at
# https://github.com/Jiansen/proxy-doctor/issues/new

Development

git clone https://github.com/Jiansen/proxy-doctor.git
cd proxy-doctor

# Install in development mode
pip install -e ".[dev,mcp]"

# Run tests
make test

# Run linter
make lint

License

MIT