fix(mcp): rediscover dead cached servers on session startup

[izumi0uu]

What does this PR do?

Fixes a stale-MCP startup bug where a dead MCPServerTask could remain parked in tools.mcp_tool._servers after keepalive/reconnect failure. Later session startup only checked name in _servers, so it skipped fresh MCP discovery and the new session silently lost that server's tools.

This change only treats a cached MCP server as already connected when it still has a live session or a live background reconnect task. If the cached task is dead and detached, startup rediscovery reconnects it and re-registers its tools.

Related Issue

Fixes #50170

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• Added _server_is_connected(...) in tools/mcp_tool.py so startup gating distinguishes live/recovering MCP tasks from dead cached entries.
• Updated register_mcp_servers() and discover_mcp_tools() to rediscover only dead cached servers, while preserving legitimate in-flight reconnect ownership.
• Added regression tests in tests/tools/test_mcp_tool.py for both:
  • dead cached server entry must be rediscovered on new session startup
  • live background reconnect must still suppress duplicate startup connects

How to Test

1. Reproduce the pre-fix state by leaving a stale MCP server entry in _servers with session=None and no live background task.
2. Run ./.venv/bin/pytest -q tests/tools/test_mcp_tool.py -k "reconnects_server_entry_that_exists_but_is_not_connected or skips_server_that_is_mid_reconnect_with_live_background_task or skips_already_connected_servers or connects_new_servers or logs_summary_on_success".
3. Confirm the new-session startup path now reconnects dead cached servers, while existing connected servers and in-flight reconnects still follow the old idempotent behavior.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/ -q and all tests pass
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform: macOS 15.x

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — or N/A
• I've updated cli-config.yaml.example if I added/changed config keys — or N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — or N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — or N/A
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A

Screenshots / Logs

Targeted validation:

• ./.venv/bin/pytest -q tests/tools/test_mcp_tool.py -k "reconnects_server_entry_that_exists_but_is_not_connected or skips_server_that_is_mid_reconnect_with_live_background_task or skips_already_connected_servers or connects_new_servers or logs_summary_on_success"
• ./.venv/bin/pytest -q tests/agent/test_turn_context.py -k "between_turns_refresh"
• ./.venv/bin/pytest -q tests/test_get_tool_definitions_cache_isolation.py

[teknium1]

Closing with credit — the best-engineered PR in this cluster. Your _server_is_connected gate (session None AND task done) plus mixed-state tests were the closest framing to the merged fix. The recent reconnect overhaul (PR #59222) made the task never-done (parks + self-probes), so the liveness half of your gate became a no-op; PR #59331 (merged) covers your issue #50170 by nudging cached dead-transport entries at session startup — credited to you in the salvage commit. Thanks!