feat: Add MCP connection keepalive/reload mechanism for CLI-37

[szmania]

Purpose

Implement a keepalive/reload mechanism for MCP server connections to prevent connection errors during periods of inactivity. Addresses Jira issue CLI-37: MCP connection keepalive/reload mechanism.

Changes

• Added ConnectionState enum to track connection status (CONNECTED, UNHEALTHY, DISCONNECTED)
• Implemented start_keepalive() and _keepalive_loop() methods to send periodic OPTIONS requests
• Added exponential backoff reconnection logic with jitter
• Added configurable keepalive interval (5–300 seconds)
• Added comprehensive test suite including:
  • Configuration validation
  • Integration tests with mock server
  • Concurrency and resilience tests
  • Logging and metrics verification

Configuration Example

The keepalive_interval can be added to your MCP server configuration. Here is an example of what an MCP object looks like in JSON format:

{
  "name": "my-mcp-server",
  "url": "http://localhost:8000",
  "type": "http",
  "keepalive_interval": 30,
  "headers": {
    "Authorization": "Bearer my-token"
  },
  "enabled": true
}

Testing

All tests pass:

• test_keepalive_config.py - Validates configuration parameters
• test_keepalive_concurrency.py - Tests task lifecycle and isolation
• test_keepalive_integration.py - Tests with mock server
• test_keepalive_logging.py - Verifies proper logging
• test_keepalive_resilience.py - Tests recovery scenarios
• test_keepalive_unit.py - Tests state transitions

Related Issues

Fixes CLI-37: MCP connection keepalive/reload mechanism

Notes

• The keepalive_interval is configurable in the server config:
  • If not provided, it defaults to 300 seconds.
  • The value must be between 5 and 300 seconds.
• Jitter (±10%) is added to prevent timing analysis.
• Reconnection uses exponential backoff (1s, 2s, 4s, 8s, 16s, 32s, capped at 300s).
• Authentication headers are preserved in keepalive requests.
• Connection state transitions are properly logged.

Checklist

• Implementation complete
• Tests added and passing
• Documentation updated
• No sensitive information in logs
• Backward compatible
• Follows existing code patterns

[szmania]

@dwash96 This one is ready

[szmania]

This addresses this error when being on cecli tui for a certain amoutn of time: File "C:\Users\<user>\AppData\Roaming\uv\tools\cecli-dev\Lib\site-packages\httpx\_transports\default.py", line 118, in map_httpcore_exceptions raise mapped_exc(message) from exc httpx.RemoteProtocolError: peer closed connection without sending complete message body (incomplete chunked read)