Auto-Reconnect for MCP Servers

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

MCP servers using SSE (Server-Sent Events) transport frequently experience connection drops due to idle timeouts (~5 minutes), requiring constant manual reconnection. This significantly disrupts development workflow when using remote MCP servers.

Specific Example:
The official Atlassian Remote MCP Server (https://mcp.atlassian.com/v1/sse) uses the deprecated SSE transport protocol, which lacks heartbeat/keep-alive messages. This causes connections to timeout after approximately 5 minutes of inactivity, resulting in:

• Failed MCP tool calls mid-session
• "Connection closed" errors
• Need to manually reconnect via /mcp menu every 5-10 minutes

This is a known issue with SSE-based MCP servers (see: triggerdotdev/trigger.dev#2134, anthropics/claude-code#3033).

Proposed Solution

Add automatic reconnection for MCP servers when connection failures are detected.

Proposed Behavior:

1. Detection: Automatically detect when an MCP server connection:

• Times out
• Returns connection errors
• Becomes unresponsive

1. Reconnection: Automatically attempt to reconnect with:

• Exponential backoff strategy
• Configurable retry attempts
• Configurable retry delays

1. User Notification (optional):

• Silent reconnection in background (preferred for known-flaky servers)
• OR notify user of reconnection attempts
• Show reconnection status in /mcp menu

Proposed Configuration:

{
  "mcpServers": {
    "Atlassian": {
      "command": "npx",
      "args": ["-y", "mcp-remote@latest", "https://mcp.atlassian.com/v1/sse"],
      "autoReconnect": {
        "enabled": true,
        "maxRetries": 10,           // -1 for infinite
        "initialRetryDelay": 2000,  // milliseconds
        "maxRetryDelay": 30000,     // milliseconds
        "backoffMultiplier": 1.5,
        "silent": true              // Don't show notifications
      }
    }
  }
}

Alternative Solutions

Current Workaround:

Users must manually:

1. Notice when operations fail
2. Run /mcp command
3. Select the disconnected server
4. Choose "Reconnect"
5. Repeat every 5-10 minutes

This interrupts flow and reduces productivity, especially when working extensively with Jira/Confluence/Compass through the Atlassian MCP.

Other Alternatives Considered:

• Wrapper scripts to restart MCP processes - doesn't work for SSE timeouts since the process doesn't exit
• Keep-alive ping mechanisms - can't be implemented client-side for remote servers
• Switching to community MCP servers - not always feasible due to security/support concerns
• Manual reconnection (current state) - significantly degrades user experience

Priority

High - Significant impact on productivity

Feature Category

MCP server integration

Use Case Example

Example scenario:

1. I'm working on a React app with a Node.js backend, using Claude Code to help with development
2. I need to frequently reference and update Jira tickets, link related issues, and update Confluence documentation
3. I have the Atlassian Remote MCP Server configured to access Jira and Confluence
4. Every 5-10 minutes, the SSE connection times out due to inactivity
5. My next attempt to use an MCP tool (e.g., "update Jira ticket CPDE-1234") fails with "Connection closed"
6. I must stop my work, run /mcp, select "Atlassian", click "Reconnect", wait for connection, then retry my command
7. This happens 6-12 times per hour during active development sessions

With auto-reconnect:

• The connection drop would be detected automatically
• Claude Code would silently reconnect in the background (2-5 seconds)
• My MCP tool call would either succeed after auto-reconnect or I'd get a clear "reconnecting..." message
• I could continue working without manual intervention
• My development flow remains uninterrupted

Additional Context

Technical Context:

• SSE (Server-Sent Events) transport was deprecated in MCP protocol version 2.0.9
• Replaced by Streamable HTTP in MCP protocol version 2025-03-26
• Microsoft dropped SSE support in Copilot Studio as of August 2025
• HTTP servers and proxies automatically close idle SSE connections (typically after 5 minutes)

Related Issues:

• #1026 - Manual reconnect feature (COMPLETED - this request is for AUTOMATIC reconnection)
• #3033 - MCP Server Timeout Configuration (related to SSE timeouts)
• triggerdotdev/trigger.dev#2134 - SSE connection drops after 5 minutes

Current Workaround Limitations:

While server-side fixes (like Atlassian upgrading from SSE to Streamable HTTP) would address specific cases, auto-reconnect provides:

• Immediate relief for users of existing flaky MCP servers
• Handles other connection issues beyond SSE limitations (network drops, VPN disconnects, corporate firewalls)
• Improves reliability for all MCP server types
• Creates a more resilient MCP integration experience overall

Affected MCP Servers:

• Atlassian Remote MCP (official)
• Trigger.dev
• Other services using SSE transport
• Any MCP server experiencing intermittent network issues

Current Claude Code Version: v1.0.64+
Severity: High - Significantly impacts daily workflow for users of remote MCP servers