Corcava logoThe Only Business Tool You NeedCorcava
Menu

Continue MCP SSE Issues: Connection Drops, Streaming, and Retries

Having trouble with Continue's SSE connection to Corcava MCP? This Continue-specific troubleshooting guide helps you fix SSE disconnects, streaming failures, and connection issues. Learn how to validate your config, check network/proxy settings, and implement recommended retry behavior.

Understanding SSE in Continue

Continue uses Server-Sent Events (SSE) for remote MCP servers. SSE provides real-time streaming but can be sensitive to network conditions:

How SSE Works in Continue

  • Persistent connection: Maintains long-lived connection to MCP server
  • Streaming responses: Server streams responses as they're generated
  • Unidirectional (server→client): SSE streams data from server to client only; clients must use separate HTTP requests or WebSockets to send data back
  • Network sensitive: More sensitive to network issues than HTTP polling

Common SSE Issues

Connection Drops

  • Symptom: SSE connection established but then drops unexpectedly
  • Causes:
    • Network instability
    • Proxy timeouts
    • Firewall closing idle connections
    • Server-side connection limits

Streaming Failures

  • Symptom: Connection works but streaming responses are incomplete or corrupted
  • Causes:
    • Proxy buffering SSE streams
    • Network packet loss
    • Client-side buffer issues

Connection Refused

  • Symptom: Cannot establish SSE connection at all
  • Causes:
    • Invalid endpoint URL
    • Network/firewall blocking SSE
    • Server not accessible
    • Authentication failure

Step 1: Validate Continue Configuration

First, ensure your Continue config is correct for SSE:

Correct SSE Config Format

Valid Continue Config

{
  "mcpServers": {
    "corcava": {
      "url": "https://app.corcava.com/mcp",
      "transport": "sse",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Key points:

  • URL must be HTTPS (SSE requires secure connection)
  • Transport should be "sse" (explicit or default)
  • Authorization header must be correctly formatted

Config Validation Checklist

  • ✅ Config file is valid JSON (see Config JSON Errors)
  • ✅ URL is correct: https://app.corcava.com/mcp
  • ✅ Transport is "sse" (or omitted, defaults to SSE for remote)
  • ✅ Authorization header has "Bearer " prefix
  • ✅ API key is active and not revoked

Step 2: Check Network and Proxy Settings

SSE connections can be blocked or interfered with by network infrastructure:

Network Diagnostics

Test SSE Endpoint

curl -N -H "Authorization: Bearer YOUR_API_KEY" \
  https://app.corcava.com/mcp

The -N flag disables buffering, useful for testing SSE streams. If this fails, there may be network/proxy issues.

Proxy Issues with SSE

Corporate Proxy Problems

Corporate proxies often interfere with SSE:

  • Buffering: Proxies may buffer SSE streams, breaking real-time behavior
  • Timeouts: Proxies may close idle SSE connections
  • Filtering: Some proxies block SSE connections entirely

Fix: Use VPN, configure proxy exceptions, or contact IT to whitelist SSE connections

Firewall Rules

Required for SSE:

  • Outbound HTTPS (port 443) to app.corcava.com
  • Long-lived connections allowed (SSE keeps connection open)
  • No connection timeout too short (SSE connections can be idle)

Step 3: Handle Connection Drops

SSE connections can drop. Implement retry logic:

Retry Strategy for SSE Drops

  1. Immediate retry: If connection drops, retry immediately (may be transient)
  2. Exponential backoff: If immediate retry fails, wait 2 seconds, then 4 seconds
  3. Max retries: 3 attempts before reporting error
  4. Reconnect: Re-establish SSE connection from scratch

Continue Automatic Retries

Continue may automatically retry SSE connections. Check Continue settings for:

  • Automatic reconnection enabled
  • Retry delay settings
  • Max retry attempts

Step 4: Verify SSE Connection

Use this simple verification workflow to test your SSE connection:

Simple Verification Workflow

  1. Check connection status: Look for SSE connection indicator in Continue
  2. Test tool listing: Ask "What MCP tools are available?"
  3. Test simple call: Try "List my projects" (read-only, fast)
  4. Monitor connection: Watch for connection drops during operation
  5. Check logs: Review Continue logs for SSE errors

Expected Behavior

When SSE is working correctly:

  • Connection indicator shows "connected" or "active"
  • Tools appear in available tools list
  • Tool calls complete successfully
  • Responses stream in real-time (if supported)
  • No connection drop errors in logs

Step 5: Troubleshoot Streaming Issues

If connection works but streaming is problematic:

Streaming Failure Symptoms

Incomplete Responses

  • Symptom: Tool calls return partial data or cut off mid-response
  • Fix: Check for proxy buffering, increase client buffer size, or use HTTP polling instead

Delayed Responses

  • Symptom: Responses arrive but with significant delay
  • Fix: Check network latency, verify proxy isn't buffering, test from different network

Corrupted Data

  • Symptom: Responses contain malformed JSON or incomplete data
  • Fix: Check for packet loss, verify network stability, try reconnecting

Step 6: Alternative: Use HTTP Polling

If SSE continues to have issues, some MCP clients support HTTP polling as an alternative:

SSE vs HTTP Polling

If SSE is unreliable in your environment:

  • HTTP polling may be more stable (though less efficient)
  • Check Continue documentation for polling transport option
  • Polling works better through corporate proxies
  • Trade-off: Less real-time but more reliable

Quick Fix Checklist

Before Escalating

  • ✅ Config file is valid JSON with correct SSE format
  • ✅ URL is correct and uses HTTPS
  • ✅ API key is active and correctly formatted
  • ✅ Tested network connection (curl test works)
  • ✅ Checked for proxy interference
  • ✅ Verified firewall allows long-lived HTTPS connections
  • ✅ Continue has been restarted after config changes
  • ✅ Checked Continue logs for specific error messages

Related Articles

401 Unauthorized in MCP: Fixing API Key and Auth Header Issues

Getting 401 Unauthorized errors when using MCP? This focused troubleshooting guide helps you fix authentication issues: missing Authorization headers, wrong Bearer format, revoked API keys, and whites...

MCP Tool Calls Fail: Debugging Inputs, Validation, and Error Payloads

MCP tools are listed but calls fail? This generic troubleshooting guide helps you debug tool call failures: invalid arguments, missing required fields, wrong IDs, and server-side validation errors. Le...

Windsurf MCP Troubleshooting: Auth Headers Not Being Sent

Windsurf connects to Corcava MCP but tool calls fail? This Windsurf-specific troubleshooting guide helps you fix cases where the server connects but tool calls fail due to missing authorization header...

Tracking Time in the Browser

You can start and stop time directly from the Corcava web app—no desktop application or API required. When you have permission to add manual time, a time tracker chip appears in the top header. Use it...

Create a Task via MCP: Required Fields, Defaults, and Safety Checks

Developer reference for the create_task MCP tool. Learn how to create new tasks with required and optional fields, understand defaults, implement safety checks, and use confirmation patterns. Includes...

List Task Comments via MCP: Timelines and Audit Trails

Developer reference for the list_task_comments MCP tool. Learn how to retrieve comment history and audit trails for tasks, including pagination and filtering. Includes example tool calls, natural lang...