feat: surface MCP server status in System Status page

Add per-tool in-memory call counters to the MCP module and a GET /api/v1/status/mcp endpoint that joins them with the live tool registry and endpoint config, then render it as a new card on the System Status page with a one-click client-config copy button. - app/mcp/stats.py: lock-guarded MCPStatsTracker (the mcp SDK runs sync tool bodies via anyio.to_thread.run_sync, so this is genuinely multi-threaded, unlike the async REST routes) - app/mcp/server.py: instrument search_regulations, add get_mcp_status() - app/config/settings.py: optional MCP_PUBLIC_URL override, required because the Vite proxy and reverse proxies rewrite the Host header - StatusPage.tsx: MCP Server card, joins the existing parallel fetch Counters are process-local by design; token usage is already persisted by ModelUsageTracker since MCP calls route through ask(). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-08-03 11:37:22 +08:00
parent 73e79a610d
commit 31bbf80aeb
14 changed files with 621 additions and 10 deletions
@@ -0,0 +1,100 @@
+"""Tests for the in-memory MCP per-tool statistics tracker.
+
+These pin the two properties the status panel depends on: counters stay exact
+under the concurrent thread dispatch the mcp SDK uses, and recording never
+raises into a live tool call.
+"""
+
+from __future__ import annotations
+
+import threading
+
+from app.mcp.stats import MCPStatsTracker, MCPToolStats, get_mcp_stats_tracker
+
+
+def test_avg_duration_is_none_before_any_call():
+    """A never-called tool reports None, not 0.0, so the UI can distinguish them."""
+    assert MCPToolStats().avg_duration_ms is None
+
+
+def test_avg_duration_is_the_mean_of_recorded_durations():
+    """Average is computed over all calls, successful or not."""
+    tracker = MCPStatsTracker()
+    for duration in (100.0, 200.0, 300.0):
+        tracker.record(tool="search_regulations", duration_ms=duration, success=True)
+
+    stats = tracker.snapshot()["search_regulations"]
+    assert stats.calls == 3
+    assert stats.avg_duration_ms == 200.0
+
+
+def test_failures_increment_both_calls_and_errors():
+    """errors is a subset of calls, so the UI can show "2 of 3 failed" honestly."""
+    tracker = MCPStatsTracker()
+    tracker.record(tool="t", duration_ms=1.0, success=True)
+    tracker.record(tool="t", duration_ms=1.0, success=False)
+    tracker.record(tool="t", duration_ms=1.0, success=False)
+
+    stats = tracker.snapshot()["t"]
+    assert stats.calls == 3
+    assert stats.errors == 2
+
+
+def test_last_called_at_is_set_and_timezone_aware():
+    """The panel renders this as a local time, which requires an aware datetime."""
+    tracker = MCPStatsTracker()
+    tracker.record(tool="t", duration_ms=1.0, success=True)
+
+    last_called = tracker.snapshot()["t"].last_called_at
+    assert last_called is not None
+    assert last_called.tzinfo is not None
+
+
+def test_concurrent_record_calls_are_not_lost():
+    """8 threads x 100 calls must total exactly 800.
+
+    Without the lock this loses increments non-deterministically. The mcp SDK
+    runs synchronous tool bodies via anyio.to_thread.run_sync, so this is the
+    real dispatch model, not a hypothetical.
+    """
+    tracker = MCPStatsTracker()
+
+    def hammer() -> None:
+        for _ in range(100):
+            tracker.record(tool="search_regulations", duration_ms=1.0, success=True)
+
+    threads = [threading.Thread(target=hammer) for _ in range(8)]
+    for thread in threads:
+        thread.start()
+    for thread in threads:
+        thread.join()
+
+    stats = tracker.snapshot()["search_regulations"]
+    assert stats.calls == 800
+    assert stats.total_duration_ms == 800.0
+
+
+def test_record_swallows_bad_input_instead_of_raising():
+    """A malformed duration must not propagate into the caller's tool call."""
+    tracker = MCPStatsTracker()
+    tracker.record(tool="t", duration_ms="not-a-number", success=True)  # type: ignore[arg-type]
+
+    # Coercion happens before the lock is taken, so the entry is never created
+    # in a half-updated state.
+    assert tracker.snapshot() == {}
+
+
+def test_snapshot_is_a_copy_not_the_live_dict():
+    """Callers mutating the snapshot must not corrupt the tracker."""
+    tracker = MCPStatsTracker()
+    tracker.record(tool="t", duration_ms=1.0, success=True)
+
+    snapshot = tracker.snapshot()
+    snapshot.clear()
+
+    assert "t" in tracker.snapshot()
+
+
+def test_get_mcp_stats_tracker_returns_a_singleton():
+    """Instrumentation and the status route must observe the same counters."""
+    assert get_mcp_stats_tracker() is get_mcp_stats_tracker()
@@ -0,0 +1,79 @@
+"""Tests for get_mcp_status(), the payload behind the System Status MCP card.
+
+Covers the join between the live tool registry and the stats tracker, plus
+the two values the route supplies or the settings decide.
+"""
+
+from __future__ import annotations
+
+import asyncio
+from unittest.mock import patch
+
+from app.mcp.stats import MCPStatsTracker
+
+
+def _status(tracker: MCPStatsTracker | None = None, **setting_overrides) -> dict:
+    """Call get_mcp_status() with an isolated tracker and patched settings.
+
+    The real tracker is a process-wide singleton, so tests must inject their
+    own instance or they leak counters into each other.
+    """
+    from app.mcp.server import get_mcp_status, settings
+
+    patched = settings.model_copy(update=setting_overrides)
+    with (
+        patch("app.mcp.server.settings", patched),
+        patch("app.mcp.server.get_mcp_stats_tracker", return_value=tracker or MCPStatsTracker()),
+    ):
+        return asyncio.run(get_mcp_status("http://6.86.80.9:8000/mcp/"))
+
+
+def test_public_url_is_passed_through_unmodified():
+    """The route owns URL resolution; get_mcp_status() must not rewrite it."""
+    assert _status()["endpoint_url"] == "http://6.86.80.9:8000/mcp/"
+
+
+def test_auth_required_follows_settings():
+    """The panel's auth badge must reflect live config, not a hard-coded value."""
+    assert _status(auth_enabled=True)["auth_required"] is True
+    assert _status(auth_enabled=False)["auth_required"] is False
+
+
+def test_allowed_hosts_are_split_and_stripped():
+    """Displayed allow-list must match the one the transport actually enforces."""
+    status = _status(mcp_allowed_hosts="6.86.80.9:* , 127.0.0.1:*,")
+    assert status["allowed_hosts"] == ["6.86.80.9:*", "127.0.0.1:*"]
+
+
+def test_tools_come_from_the_live_registry_with_zeroed_stats():
+    """An advertised but never-called tool reports zeros, not absence."""
+    tools = {tool["name"]: tool for tool in _status()["tools"]}
+
+    assert "search_regulations" in tools
+    assert tools["search_regulations"]["calls"] == 0
+    assert tools["search_regulations"]["errors"] == 0
+    assert tools["search_regulations"]["avg_duration_ms"] is None
+    assert tools["search_regulations"]["last_called_at"] is None
+
+
+def test_recorded_stats_are_joined_onto_the_matching_tool():
+    """Counters recorded by the instrumented tool must surface on that tool's row."""
+    tracker = MCPStatsTracker()
+    tracker.record(tool="search_regulations", duration_ms=120.0, success=True)
+    tracker.record(tool="search_regulations", duration_ms=80.0, success=False)
+
+    tool = next(t for t in _status(tracker)["tools"] if t["name"] == "search_regulations")
+
+    assert tool["calls"] == 2
+    assert tool["errors"] == 1
+    assert tool["avg_duration_ms"] == 100.0
+    # Serialized for JSON transport; the frontend parses it with new Date().
+    assert isinstance(tool["last_called_at"], str)
+
+
+def test_description_is_the_first_docstring_line():
+    """Multi-line tool docstrings must not blow up the card's row height."""
+    tool = next(t for t in _status()["tools"] if t["name"] == "search_regulations")
+
+    assert "\n" not in tool["description"]
+    assert tool["description"].startswith("Search the compliance knowledge base")