fix(mcp): don't trip circuit breaker on tool errors

[renga-kogahara]

What does this PR do?

Treats completed MCP CallToolResult(isError=True) responses as tool/domain errors rather than server connectivity failures.

A completed RPC proves the MCP transport is reachable. Counting a tool-level validation error toward the server circuit breaker can incorrectly mark a healthy MCP server as unreachable and prevent subsequent valid calls.

Related Issue

Supersedes #74718.

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

• Reset the server connectivity breaker after any completed tools/call RPC, including isError=True results.
• Preserve completed tool/domain errors returned after OAuth recovery or session reconnect instead of reclassifying them as auth/transport failures.
• Continue counting raised timeout, transport, and session failures toward the connectivity breaker.
• Add regression coverage for normal calls, OAuth retries, and session-reconnect retries.

How to Test

1. Run .venv/bin/python -m pytest tests/tools/test_mcp*.py -q -o 'addopts='.
2. Confirm repeated CallToolResult(isError=True) responses do not open the server connectivity breaker.
3. Confirm OAuth and session reconnect retries return completed tool errors unchanged.

Focused result: 385 passed.

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: Linux

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — docstrings updated
• I've updated cli-config.yaml.example if I added/changed config keys — N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — no platform-specific APIs added
• I've updated tool descriptions/schemas if I changed tool behavior — N/A

Screenshots / Logs

N/A

[alt-glitch]

This was generated by AI during triage.

Duplicate of #74045. Both reset the MCP connectivity breaker after a completed tool/domain-error RPC; #74045 also covers the auth and session-recovery paths, so this narrower patch is redundant.

[teknium1]

Thanks for the focused MCP classification fix. Current main still converts completed CallToolResult.isError responses into an error envelope at tools/mcp_tool.py:4812-4827 and then increments the server-wide breaker for that envelope at tools/mcp_tool.py:4896-4905, so the reported behavior is present.

Problems

• This patch is redundant with the still-open #74045: its production changes cover the same auth retry, session-reconnect retry, and completed-call classification sites. #74045 also exercises recovery through _make_tool_handler, rather than only calling the shared retry helpers directly.

Suggested changes

• If retained separately, test the auth and session recovery paths through _make_tool_handler with a real CallToolResult(isError=True) fixture, covering the conversion path at tools/mcp_tool.py:4812-4827 as well as the recovery helper.

This is an automated hermes-sweeper review.

[teknium1]

This invokes the private recovery helper with a prebuilt JSON error, so it does not exercise the real CallToolResult.isError → tool_error(...) conversion in _make_tool_handler. Consider a handler-level regression like #74045's recovery coverage.

[teknium1]

This likewise verifies the helper return value but not the actual post-reconnect CallToolResult.isError conversion path. A handler-level fixture would cover the end-to-end behavior this PR changes.