diff --git a/src/skillspector/mcp_server.py b/src/skillspector/mcp_server.py
index 444b75f..f1119bd 100644
--- a/src/skillspector/mcp_server.py
+++ b/src/skillspector/mcp_server.py
@@ -33,7 +33,7 @@
 from skillspector import __version__
 from skillspector.graph import graph
 from skillspector.logging_config import get_logger
-from skillspector.providers import resolve_provider_credentials
+from skillspector.providers import resolve_chat_model_credentials
 
 if TYPE_CHECKING:
     from mcp.server.fastmcp import FastMCP
@@ -74,7 +74,13 @@ async def run_scan(
     if output_format not in VALID_FORMATS:
         raise ValueError(f"output_format must be one of {VALID_FORMATS}, got {output_format!r}")
 
-    llm_available = resolve_provider_credentials() is not None
+    # Use the chat-model credential resolver, not the active-provider-only one,
+    # so availability matches what ``create_chat_model`` actually does: it falls
+    # back to a standard OpenAI client (OPENAI_API_KEY / OPENAI_BASE_URL) when the
+    # active provider is unconfigured. Gating on the active provider alone made the
+    # server report llm_available=false and skip the LLM pass for an OpenAI-only
+    # setup the CLI would have run (see issue #200).
+    llm_available = resolve_chat_model_credentials() is not None
     llm_used = use_llm and llm_available
 
     state: dict[str, Any] = {
diff --git a/tests/unit/test_mcp_server.py b/tests/unit/test_mcp_server.py
index 10c5596..b61751e 100644
--- a/tests/unit/test_mcp_server.py
+++ b/tests/unit/test_mcp_server.py
@@ -33,7 +33,7 @@ async def test_run_scan_returns_structured_verdict(
 ) -> None:
     """run_scan returns a JSON-serialisable verdict with the expected shape."""
     # No credentials: the LLM pass cannot run regardless of what is requested.
-    monkeypatch.setattr(mcp_server, "resolve_provider_credentials", lambda: None)
+    monkeypatch.setattr(mcp_server, "resolve_chat_model_credentials", lambda: None)
     _write_skill(tmp_path)
 
     result = await run_scan(str(tmp_path), use_llm=True, output_format="json")
@@ -51,7 +51,7 @@ async def test_run_scan_llm_accounting_is_honest_without_credentials(
     tmp_path: Path, monkeypatch: pytest.MonkeyPatch
 ) -> None:
     """Requesting the LLM with no credentials must report it as not used."""
-    monkeypatch.setattr(mcp_server, "resolve_provider_credentials", lambda: None)
+    monkeypatch.setattr(mcp_server, "resolve_chat_model_credentials", lambda: None)
     _write_skill(tmp_path)
 
     result = await run_scan(str(tmp_path), use_llm=True, output_format="json")
@@ -66,7 +66,7 @@ async def test_run_scan_reports_llm_available_with_credentials(
     tmp_path: Path, monkeypatch: pytest.MonkeyPatch
 ) -> None:
     """Credentials present but use_llm=False: available, but honestly not used."""
-    monkeypatch.setattr(mcp_server, "resolve_provider_credentials", lambda: ("key", None))
+    monkeypatch.setattr(mcp_server, "resolve_chat_model_credentials", lambda: ("key", None))
     _write_skill(tmp_path)
 
     result = await run_scan(str(tmp_path), use_llm=False, output_format="json")
@@ -77,6 +77,29 @@ async def test_run_scan_reports_llm_available_with_credentials(
     assert result["scan_mode"] == "static-only"
 
 
+async def test_run_scan_reports_llm_available_via_openai_fallback(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """Regression for #200: availability follows the chat-model resolver.
+
+    When the active provider is unconfigured but the standard OpenAI fallback
+    (OPENAI_API_KEY / OPENAI_BASE_URL) is set, ``create_chat_model`` would run the
+    LLM, so ``resolve_chat_model_credentials`` resolves credentials even though
+    ``resolve_provider_credentials`` (active provider only) returns ``None``. The
+    server must report the LLM as available and must not gate it out. Patching the
+    active-provider-only resolver to ``None`` proves the gate no longer consults it.
+    """
+    monkeypatch.setattr(mcp_server, "resolve_chat_model_credentials", lambda: ("openai-key", None))
+    if hasattr(mcp_server, "resolve_provider_credentials"):
+        monkeypatch.setattr(mcp_server, "resolve_provider_credentials", lambda: None)
+    _write_skill(tmp_path)
+
+    result = await run_scan(str(tmp_path), use_llm=False, output_format="json")
+
+    assert result["llm_available"] is True
+    assert result["scan_mode"] == "static-only"  # use_llm=False, so still not used
+
+
 async def test_run_scan_rejects_invalid_format(tmp_path: Path) -> None:
     """An unsupported output_format is rejected before any scan runs."""
     with pytest.raises(ValueError):