Can’t Connect to Cursor AI Server? Complete Troubleshooting Guide

Cursor

Check first

Anthropic (Claude)

Cursor uses this

OpenAI (GPT-4o)

Cursor uses this

Cursor routes AI requests through both Anthropic and OpenAI. If either provider is degraded, Cursor AI features may fail even when Cursor’s own infrastructure is Operational.

Step 1: Check if Cursor is down

Before changing any settings, rule out a service-wide outage. If Cursor’s backend is down, no local fix will work — you just have to wait.

Step 2: Sign out and re-authenticate

Cursor uses short-lived session tokens. When one expires, the AI backend rejects requests with “Authentication failed” or a generic connection error. This is the fix for most users who were connected yesterday but aren’t today.

1. Open Cursor Settings → Account
2. Click Sign Out and confirm
3. Sign back in with the same account
4. Press Ctrl+Shift+P → Reload Window
5. Verify your subscription is Active at cursor.com/settings

Step 3: Fix “Invalid API key” in Bring-Your-Own-Key mode

If you configured Cursor to use your own OpenAI or Anthropic API key (BYOK mode), the “Invalid API key” error means the key is wrong, expired, or lacks credit. Steps to fix:

1. Go to Cursor Settings → Models
2. Delete the existing key and generate a fresh one from your provider’s dashboard
3. For OpenAI: platform.openai.com/api-keys — confirm billing is active
4. For Anthropic: console.anthropic.com/settings/keys — confirm you have remaining credits
5. Paste the new key and click Verify

You can quickly test an Anthropic key from the terminal:

# Test Anthropic key validity
curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: YOUR_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-3-haiku-20240307","max_tokens":16,"messages":[{"role":"user","content":"ping"}]}'

# Expected: {"id":"msg_...","type":"message",...}
# Error: {"type":"error","error":{"type":"authentication_error",...}}

# Test OpenAI key validity
curl https://api.openai.com/v1/models \
  -H "Authorization: Bearer YOUR_API_KEY"

# Expected: {"object":"list","data":[...]}
# Error 401: {"error":{"message":"Incorrect API key provided",...}}

Step 4: Fix HTTP/2 proxy incompatibility

Cursor uses HTTP/2 for streaming AI responses. Many corporate proxies — including older Squid and Zscaler deployments — do not support HTTP/2 and silently drop the connections, producing “connection timeout” or “stream error” messages.

Force Cursor to use HTTP/1.1 by adding this to your settings.json (Ctrl+Shift+P → Open User Settings (JSON)):

{
  "cursor.http.httpVersion": "http1.1",
  "cursor.general.gitStrictHostKeyChecking": false
}

After saving, reload the window (Ctrl+Shift+P → Reload Window). If Cursor then connects, your proxy was the culprit. Report the domain to your IT team so they can add an HTTP/2 exception for *.cursor.sh.

Step 5: Allowlist Cursor domains in your firewall or proxy

Corporate firewalls, DNS filters (Cisco Umbrella, Cloudflare Gateway), and VPNs frequently block AI tool domains by category. Share this list with your IT administrator:

# Cursor required domains — add to proxy/firewall allowlist
cursor_domains:
  - cursor.sh
  - "*.cursor.sh"          # covers api.cursor.sh, auth.cursor.sh, etc.
  - cursor-cdn.anysphere.inc

# AI model backends Cursor routes through
ai_backends:
  - api.anthropic.com      # Claude models
  - api.openai.com         # GPT-4o, GPT-4-turbo
  - "*.openai.com"

# Authentication and telemetry
auth:
  - auth0.cursor.sh
  - sentry.io              # optional: crash reporting

If your proxy requires an explicit HTTPS_PROXY setting, configure it via the environment variable before launching Cursor:

# Windows — set before launching Cursor
set HTTPS_PROXY=http://proxy.company.com:8080
set HTTP_PROXY=http://proxy.company.com:8080
set NO_PROXY=localhost,127.0.0.1

# macOS / Linux
export HTTPS_PROXY=http://proxy.company.com:8080
export HTTP_PROXY=http://proxy.company.com:8080
export NO_PROXY=localhost,127.0.0.1

# Then launch Cursor from the same terminal session
cursor .

export NODE_EXTRA_CA_CERTS=/path/to/corporate-root-ca.pem

Step 6: Run Cursor’s built-in network diagnostics

Cursor ships a diagnostic tool that tests every endpoint it depends on and outputs a structured report — useful for a support ticket or for showing your IT team exactly what is blocked.

1. Inside Cursor, open the Command Palette: Ctrl+Shift+P (Windows/Linux) or Cmd+Shift+P (Mac)
2. Type Network Diagnostics and select Cursor: Run Network Diagnostics
3. A panel opens showing ping results for each domain, HTTP status codes, and latency
4. Any domain showing FAILED or TIMEOUT is being blocked

You can also run a quick connectivity test manually to cross-check:

# Test Cursor backend reachability
curl -v --max-time 10 https://api.cursor.sh/health 2>&1 | grep -E "Connected|SSL|HTTP|curl"

# Test Anthropic backend
curl -I --max-time 10 https://api.anthropic.com

# Test OpenAI backend
curl -I --max-time 10 https://api.openai.com

# Check if HTTP/2 is being used
curl -v --http2 https://api.cursor.sh 2>&1 | grep "Using HTTP"

Last resort: switch to Bring-Your-Own-Key mode

If Cursor’s own backend is still blocked after allowlisting, you can bypass it by providing your own API key directly. Cursor then contacts Anthropic or OpenAI directly — which may be on a different network path your firewall allows.

1. Cursor Settings → Models
2. Toggle OpenAI API Key or Anthropic API Key
3. Paste your key and click Verify
4. Select a model your key has access to (e.g., claude-3-5-sonnet-20241022)

Note: BYOK mode bills against your own API account, separate from your Cursor subscription.

Frequently asked questions