Implement MCP OAuth 2.0/2.1 authentication with Dynamic Client Registration (fixes 401 on protected servers like Notion)

Formula/omlx.rb

@@ -1,8 +1,8 @@
 class Omlx < Formula
   desc "LLM inference server optimized for Apple Silicon"
   homepage "https://github.com/jundot/omlx"
-  url "https://github.com/jundot/omlx/archive/refs/tags/v0.3.5.tar.gz"
-  sha256 "d40f7b13a35e944f0c00fd9005e6667bd2b8be083f6cce3396f97a733fe22876"
+  url "https://github.com/jundot/omlx/archive/refs/tags/v0.3.6.tar.gz"
+  sha256 "61135fcc60ca7f9b2a9da3d6c06646963a374f9173918d484916933636ab058b"
   license "Apache-2.0"
 
   head "https://github.com/jundot/omlx.git", branch: "main"

conftest.py

@@ -0,0 +1,55 @@
+# SPDX-License-Identifier: Apache-2.0
+"""
+Root conftest.py — mock Apple-Silicon-only dependencies so that unit tests
+for pure-Python modules (e.g. omlx.mcp.*) can run on Linux CI runners.
+"""
+
+import sys
+import types
+from unittest.mock import MagicMock
+
+
+def _mock_mlx() -> None:
+    """Insert MagicMock stubs for mlx and mlx_lm before any imports occur."""
+    if "mlx.core" in sys.modules:
+        # Already importable (macOS with MLX installed) — nothing to do.
+        try:
+            import mlx.core  # noqa: F401
+            return
+        except ImportError:
+            pass
+
+    # Build a minimal package tree that satisfies all sub-module imports
+    # encountered in omlx/scheduler.py and friends.
+    for pkg_name in ("mlx", "mlx_lm", "mlx_embeddings", "mlx_vlm"):
+        pkg = types.ModuleType(pkg_name)
+        pkg.__path__ = []  # mark as package
+        sys.modules[pkg_name] = pkg
+
+    # Attach commonly referenced sub-modules
+    _submodules = [
+        "mlx.core",
+        "mlx.nn",
+        "mlx.nn.layers",
+        "mlx.optimizers",
+        "mlx.utils",
+        "mlx_lm.generate",
+        "mlx_lm.utils",
+        "mlx_lm.models",
+        "mlx_lm.models.base",
+        "mlx_lm.models.cache",
+        "mlx_lm.sample_utils",
+        "mlx_lm.tokenizer_utils",
+        "mlx_embeddings.core",
+        "mlx_vlm.utils",
+    ]
+    for name in _submodules:
+        if name not in sys.modules:
+            sys.modules[name] = MagicMock()
+
+    # Make mlx.core attributes look real enough for isinstance checks
+    mx = sys.modules["mlx.core"]
+    mx.array = MagicMock  # type: ignore[attr-defined]
+
+
+_mock_mlx()

mcp.example.json

@@ -19,6 +19,30 @@
       "url": "http://localhost:3001/sse",
       "enabled": false,
       "timeout": 60
+    },
+    "notion": {
+      "_comment": "Notion remote MCP — uses OAuth 2.1 with Dynamic Client Registration (no client_id needed). Run: omlx mcp login notion",
+      "transport": "streamable-http",
+      "url": "https://mcp.notion.com/mcp",
+      "enabled": false,
+      "timeout": 60,
+      "auth": {
+        "type": "oauth2"
+      }
+    },
+    "notion-oauth-app": {
+      "_comment": "Notion public OAuth app — for developers with a registered OAuth application (client_id required)",
+      "transport": "streamable-http",
+      "url": "https://mcp.notion.com/mcp",
+      "enabled": false,
+      "timeout": 60,
+      "auth": {
+        "type": "oauth2",
+        "client_id": "YOUR_NOTION_OAUTH_CLIENT_ID",
+        "auth_url": "https://api.notion.com/v1/oauth/authorize",
+        "token_url": "https://api.notion.com/v1/oauth/token",
+        "scopes": ["read_content", "update_content"]
+      }
     }
   },
   "max_tool_calls": 10,

omlx/_version.py

@@ -1 +1 @@
-__version__ = "0.3.5"
+__version__ = "0.3.6"

omlx/adapter/gemma4.py

@@ -205,14 +205,15 @@ def extract_gemma4_messages(
                 i += 1
 
             if tool_responses:
-                processed.append(
-                    {
-                        "role": "assistant",
-                        "content": "",
-                        "tool_responses": tool_responses,
-                        _PRESERVE_BOUNDARY_KEY: True,
-                    }
-                )
+                # Attach tool_responses to the SAME assistant message that
+                # has tool_calls.  The Gemma 4 chat template checks for
+                # tool_responses on the current message (lines 261-267)
+                # BEFORE falling back to a forward-scan for role='tool'
+                # messages (lines 268-302).  Putting them on a separate
+                # assistant message causes both paths to miss, producing a
+                # corrupt bare <|tool_response> tag and making the model
+                # loop on the same tool call.
+                out_msg["tool_responses"] = tool_responses
             continue
 
         # All other roles (user, system)

omlx/admin/accuracy_benchmark.py

@@ -33,9 +33,10 @@
 _engine_pool_ref: Any = None
 
 VALID_BENCHMARKS = [
-    "mmlu", "kmmlu", "cmmlu", "jmmlu",
+    "mmlu", "mmlu_pro", "kmmlu", "cmmlu", "jmmlu",
     "hellaswag", "truthfulqa", "arc_challenge", "winogrande",
-    "gsm8k", "humaneval", "mbpp", "livecodebench",
+    "gsm8k", "mathqa", "humaneval", "mbpp", "livecodebench",
+    "bbq", "safetybench",
 ]
 
 
@@ -440,6 +441,7 @@ async def on_progress(current: int, total: int) -> None:
                         "predicted": qr.predicted,
                         "question": qr.question_text,
                         "raw_response": qr.raw_response,
+                        "category": qr.category,
                         "time_s": round(qr.time_seconds, 3),
                     }
                     for qr in result.question_results

omlx/admin/i18n/en.json

@@ -196,6 +196,8 @@
   "models.oq.cancel_tooltip": "Cancel quantization",
   "models.oq.remove_tooltip": "Remove from list",
   "models.oq.advanced_settings": "Advanced Settings",
+  "models.oq.dtype_label": "Non-quant weight dtype",
+  "models.oq.dtype_help": "float16 gives ~20% faster prefill on M1/M2 Apple Silicon (native fp16). bfloat16 is safer on M3/M4 and for numerical stability. Output name appends '-fp16' when float16 is selected.",
   "models.uploader.section_label": "Hub Upload",
   "models.uploader.heading": "Upload oQ Models",
   "models.uploader.description": "Upload your locally quantized oQ models to HuggingFace Hub.",
@@ -428,6 +430,26 @@
   "modal.model_settings.cancel": "Cancel",
   "modal.model_settings.save": "Save",
   "modal.model_settings.saving": "Saving...",
+  "modal.model_settings.profiles.section_label": "Profiles",
+  "modal.model_settings.profiles.templates_row": "Templates (global)",
+  "modal.model_settings.profiles.profiles_row": "Profiles (this model)",
+  "modal.model_settings.profiles.new_template": "+ New template",
+  "modal.model_settings.profiles.new_profile": "+ New profile",
+  "modal.model_settings.profiles.name_label": "Name",
+  "modal.model_settings.profiles.name_placeholder": "ID e.g. coding-v1",
+  "modal.model_settings.profiles.display_name_label": "Display name",
+  "modal.model_settings.profiles.description_label": "Description (optional)",
+  "modal.model_settings.profiles.also_as_template": "Also save as global template",
+  "modal.model_settings.profiles.save": "Save",
+  "modal.model_settings.profiles.cancel": "Cancel",
+  "modal.model_settings.profiles.update_from_form": "Update to current form values",
+  "modal.model_settings.profiles.resync_template": "Re-sync from template",
+  "modal.model_settings.profiles.delete": "Delete",
+  "modal.model_settings.profiles.delete_confirm": "Delete this profile?",
+  "modal.model_settings.profiles.drift_hint": "Form has unsaved changes vs. profile",
+  "modal.model_settings.profiles.custom_label": "custom",
+  "modal.model_settings.profiles.invalid_name": "Name must be lowercase letters, digits, underscore, or dash (1-32 chars)",
+  "models.list.profile_chip": "Profile",
   "logs.section_label": "Logs",
   "logs.heading": "Server Logs",
   "logs.description": "View real-time server logs with auto-refresh.",
@@ -589,6 +611,8 @@
   "chat.error.invalid_image_type": "Please select an image file",
   "chat.error.image_too_large": "Image size must be less than 10MB",
   "chat.error.image_load_failed": "Failed to load image. Please try again.",
+  "chat.tools_enabled": "Tools enabled",
+  "chat.tools_disabled": "Tools disabled",
   "js.error.required_fields": "Required fields cannot be empty: {fields}",
   "js.error.api_key_min_length": "API key must be at least 4 characters",
   "js.error.api_key_no_whitespace": "API key must not contain whitespace",

omlx/admin/i18n/ja.json

@@ -196,6 +196,8 @@
   "models.oq.cancel_tooltip": "量子化をキャンセル",
   "models.oq.remove_tooltip": "リストから削除",
   "models.oq.advanced_settings": "詳細設定",
+  "models.oq.dtype_label": "Non-quant weight dtype",
+  "models.oq.dtype_help": "float16 gives ~20% faster prefill on M1/M2 Apple Silicon (native fp16). bfloat16 is safer on M3/M4 and for numerical stability. Output name appends '-fp16' when float16 is selected.",
   "models.uploader.section_label": "Hub Upload",
   "models.uploader.heading": "Upload oQ Models",
   "models.uploader.description": "Upload your locally quantized oQ models to HuggingFace Hub.",
@@ -428,6 +430,26 @@
   "modal.model_settings.cancel": "キャンセル",
   "modal.model_settings.save": "保存",
   "modal.model_settings.saving": "保存中...",
+  "modal.model_settings.profiles.section_label": "プロファイル",
+  "modal.model_settings.profiles.templates_row": "テンプレート（グローバル）",
+  "modal.model_settings.profiles.profiles_row": "プロファイル（このモデル）",
+  "modal.model_settings.profiles.new_template": "+ 新規テンプレート",
+  "modal.model_settings.profiles.new_profile": "+ 新規プロファイル",
+  "modal.model_settings.profiles.name_label": "名前",
+  "modal.model_settings.profiles.name_placeholder": "ID 例: coding-v1",
+  "modal.model_settings.profiles.display_name_label": "表示名",
+  "modal.model_settings.profiles.description_label": "説明（任意）",
+  "modal.model_settings.profiles.also_as_template": "グローバルテンプレートとしても保存",
+  "modal.model_settings.profiles.save": "保存",
+  "modal.model_settings.profiles.cancel": "キャンセル",
+  "modal.model_settings.profiles.update_from_form": "現在のフォーム値で更新",
+  "modal.model_settings.profiles.resync_template": "テンプレートから再同期",
+  "modal.model_settings.profiles.delete": "削除",
+  "modal.model_settings.profiles.delete_confirm": "このプロファイルを削除しますか？",
+  "modal.model_settings.profiles.drift_hint": "フォームにプロファイルと異なる未保存の変更があります",
+  "modal.model_settings.profiles.custom_label": "カスタム",
+  "modal.model_settings.profiles.invalid_name": "名前は小文字・数字・アンダースコア・ハイフンのみ（1-32 文字）",
+  "models.list.profile_chip": "プロファイル",
   "logs.section_label": "ログ",
   "logs.heading": "サーバーログ",
   "logs.description": "自動更新でリアルタイムのサーバーログを確認します。",

omlx/admin/i18n/ko.json

@@ -196,6 +196,8 @@
   "models.oq.cancel_tooltip": "양자화 취소",
   "models.oq.remove_tooltip": "목록에서 제거",
   "models.oq.advanced_settings": "고급 설정",
+  "models.oq.dtype_label": "Non-quant weight dtype",
+  "models.oq.dtype_help": "float16 gives ~20% faster prefill on M1/M2 Apple Silicon (native fp16). bfloat16 is safer on M3/M4 and for numerical stability. Output name appends '-fp16' when float16 is selected.",
   "models.uploader.section_label": "Hub Upload",
   "models.uploader.heading": "Upload oQ Models",
   "models.uploader.description": "Upload your locally quantized oQ models to HuggingFace Hub.",
@@ -428,6 +430,26 @@
   "modal.model_settings.cancel": "취소",
   "modal.model_settings.save": "저장",
   "modal.model_settings.saving": "저장 중...",
+  "modal.model_settings.profiles.section_label": "프로필",
+  "modal.model_settings.profiles.templates_row": "템플릿 (전역)",
+  "modal.model_settings.profiles.profiles_row": "프로필 (이 모델)",
+  "modal.model_settings.profiles.new_template": "+ 새 템플릿",
+  "modal.model_settings.profiles.new_profile": "+ 새 프로필",
+  "modal.model_settings.profiles.name_label": "이름",
+  "modal.model_settings.profiles.name_placeholder": "ID 예: coding-v1",
+  "modal.model_settings.profiles.display_name_label": "표시 이름",
+  "modal.model_settings.profiles.description_label": "설명 (선택)",
+  "modal.model_settings.profiles.also_as_template": "전역 템플릿으로도 저장",
+  "modal.model_settings.profiles.save": "저장",
+  "modal.model_settings.profiles.cancel": "취소",
+  "modal.model_settings.profiles.update_from_form": "현재 폼 값으로 업데이트",
+  "modal.model_settings.profiles.resync_template": "템플릿에서 다시 동기화",
+  "modal.model_settings.profiles.delete": "삭제",
+  "modal.model_settings.profiles.delete_confirm": "이 프로필을 삭제하시겠습니까?",
+  "modal.model_settings.profiles.drift_hint": "폼에 프로필과 다른 저장되지 않은 변경사항이 있습니다",
+  "modal.model_settings.profiles.custom_label": "사용자 지정",
+  "modal.model_settings.profiles.invalid_name": "이름은 소문자, 숫자, 밑줄, 하이픈만 허용 (1-32자)",
+  "models.list.profile_chip": "프로필",
   "logs.section_label": "로그",
   "logs.heading": "서버 로그",
   "logs.description": "실시간 서버 로그를 자동 새로고침으로 확인합니다.",

omlx/admin/i18n/zh-TW.json

@@ -196,6 +196,8 @@
   "models.oq.cancel_tooltip": "取消量化",
   "models.oq.remove_tooltip": "從列表中移除",
   "models.oq.advanced_settings": "進階設定",
+  "models.oq.dtype_label": "Non-quant weight dtype",
+  "models.oq.dtype_help": "float16 gives ~20% faster prefill on M1/M2 Apple Silicon (native fp16). bfloat16 is safer on M3/M4 and for numerical stability. Output name appends '-fp16' when float16 is selected.",
   "models.uploader.section_label": "Hub Upload",
   "models.uploader.heading": "Upload oQ Models",
   "models.uploader.description": "Upload your locally quantized oQ models to HuggingFace Hub.",
@@ -428,6 +430,26 @@
   "modal.model_settings.cancel": "取消",
   "modal.model_settings.save": "儲存",
   "modal.model_settings.saving": "儲存中...",
+  "modal.model_settings.profiles.section_label": "設定檔",
+  "modal.model_settings.profiles.templates_row": "範本（全域）",
+  "modal.model_settings.profiles.profiles_row": "設定檔（目前模型）",
+  "modal.model_settings.profiles.new_template": "+ 新增範本",
+  "modal.model_settings.profiles.new_profile": "+ 新增設定檔",
+  "modal.model_settings.profiles.name_label": "名稱",
+  "modal.model_settings.profiles.name_placeholder": "ID，例如 coding-v1",
+  "modal.model_settings.profiles.display_name_label": "顯示名稱",
+  "modal.model_settings.profiles.description_label": "說明（選填）",
+  "modal.model_settings.profiles.also_as_template": "同時儲存為全域範本",
+  "modal.model_settings.profiles.save": "儲存",
+  "modal.model_settings.profiles.cancel": "取消",
+  "modal.model_settings.profiles.update_from_form": "更新為目前表單值",
+  "modal.model_settings.profiles.resync_template": "從範本重新同步",
+  "modal.model_settings.profiles.delete": "刪除",
+  "modal.model_settings.profiles.delete_confirm": "刪除此設定檔？",
+  "modal.model_settings.profiles.drift_hint": "表單相對設定檔有未儲存的變動",
+  "modal.model_settings.profiles.custom_label": "自訂",
+  "modal.model_settings.profiles.invalid_name": "名稱僅允許小寫字母、數字、底線、連字號（1-32 字元）",
+  "models.list.profile_chip": "設定檔",
   "logs.section_label": "日誌",
   "logs.heading": "伺服器 Logs",
   "logs.description": "檢視即時伺服器 Logs，支援自動重新整理。",

omlx/admin/i18n/zh.json

@@ -196,6 +196,8 @@
   "models.oq.cancel_tooltip": "取消量化",
   "models.oq.remove_tooltip": "从列表中移除",
   "models.oq.advanced_settings": "高级设置",
+  "models.oq.dtype_label": "Non-quant weight dtype",
+  "models.oq.dtype_help": "float16 gives ~20% faster prefill on M1/M2 Apple Silicon (native fp16). bfloat16 is safer on M3/M4 and for numerical stability. Output name appends '-fp16' when float16 is selected.",
   "models.uploader.section_label": "Hub Upload",
   "models.uploader.heading": "Upload oQ Models",
   "models.uploader.description": "Upload your locally quantized oQ models to HuggingFace Hub.",
@@ -428,6 +430,26 @@
   "modal.model_settings.cancel": "取消",
   "modal.model_settings.save": "保存",
   "modal.model_settings.saving": "保存中...",
+  "modal.model_settings.profiles.section_label": "配置档",
+  "modal.model_settings.profiles.templates_row": "模板（全局）",
+  "modal.model_settings.profiles.profiles_row": "配置档（当前模型）",
+  "modal.model_settings.profiles.new_template": "+ 新建模板",
+  "modal.model_settings.profiles.new_profile": "+ 新建配置档",
+  "modal.model_settings.profiles.name_label": "名称",
+  "modal.model_settings.profiles.name_placeholder": "ID，例如 coding-v1",
+  "modal.model_settings.profiles.display_name_label": "显示名",
+  "modal.model_settings.profiles.description_label": "描述（可选）",
+  "modal.model_settings.profiles.also_as_template": "同时保存为全局模板",
+  "modal.model_settings.profiles.save": "保存",
+  "modal.model_settings.profiles.cancel": "取消",
+  "modal.model_settings.profiles.update_from_form": "更新为当前表单值",
+  "modal.model_settings.profiles.resync_template": "从模板重新同步",
+  "modal.model_settings.profiles.delete": "删除",
+  "modal.model_settings.profiles.delete_confirm": "删除此配置档？",
+  "modal.model_settings.profiles.drift_hint": "表单相对配置档有未保存的改动",
+  "modal.model_settings.profiles.custom_label": "自定义",
+  "modal.model_settings.profiles.invalid_name": "名称仅允许小写字母、数字、下划线、短横线（1-32 字符）",
+  "models.list.profile_chip": "配置档",
   "logs.section_label": "日志",
   "logs.heading": "服务器日志",
   "logs.description": "查看实时服务器日志，支持自动刷新。",