diff --git a/app/api/endpoints/media.py b/app/api/endpoints/media.py
index ef1e67c4f..3782eade0 100644
--- a/app/api/endpoints/media.py
+++ b/app/api/endpoints/media.py
@@ -271,9 +271,8 @@ async def recognize(
         metainfo,
         media_source=media_source,
     )
-    if mediainfo:
-        return Context(meta_info=metainfo, media_info=mediainfo).to_dict()
-    return _SchemaContext()
+    # 未匹配到媒体时也返回解析元数据，便于识别测试查看识别词应用情况和解析结果
+    return Context(meta_info=metainfo, media_info=mediainfo).to_dict()
 
 
 @router.get(
diff --git a/docs/mcp-api.md b/docs/mcp-api.md
index 905479980..46540a131 100644
--- a/docs/mcp-api.md
+++ b/docs/mcp-api.md
@@ -443,7 +443,7 @@ FastAPI 的 HTTP 异常和参数校验异常统一使用 `message`，不再返
 | 方法 | 路径 | 说明 |
 | :--- | :--- | :--- |
 | GET | `/api/v1/media/search` | 按标题搜索媒体、合集、影视人物或音乐；人物结果可同时包含 MusicBrainz 艺术家，参数：`title`、`type`、`page`、`count`，可重复传入可选 `media_source`；未指定 `music_type` 时音乐搜索只返回单曲和专辑，显式 `music_type` 仍可用于专用实体选择；内置模块只处理自身支持的来源，插件模块可以处理其注册的扩展来源，旧客户端的逗号格式仅在输入边界兼容 |
-| GET | `/api/v1/media/recognize` | 识别标题，参数：`title`、`subtitle`、`custom_words`，可选 `media_source`；当 `title` 为含目录的媒体文件路径时，会合并父目录中的名称、年份等信息 |
+| GET | `/api/v1/media/recognize` | 识别标题，参数：`title`、`subtitle`、`custom_words`，可选 `media_source`；当 `title` 为含目录的媒体文件路径时，会合并父目录中的名称、年份等信息；未匹配到媒体时仍返回解析元数据 `meta_info`（含实际应用的识别词 `apply_words`），`media_info` 为空 |
 | GET | `/api/v1/media/recognize_file` | 识别文件路径，参数：`path`，可选 `media_source` |
 | GET | `/api/v1/media/{media_id}` | 按原生 ID 查询影视或音乐详情；必填参数：`media_source`、`type_name`，其中 `media_source` 与路径中的 `media_id` 组成统一媒体身份，`type_name` 支持电影、电视剧和音乐 |
 | POST | `/api/v1/media/scrape/{storage}` | 刮削媒体元数据；请求体为 `FileItem`，可选查询参数 `media_source`、`media_id`、`type_name`（电影/电视剧/音乐）和 TMDB `episode_group`。剧集组用于指定电视剧的季集顺序；音乐会按策略处理音频标签、封面和歌词。会覆盖目标位置已有的图片、NFO 和音频标签，与文件管理、手动整理一样需要管理权限 |
diff --git a/scripts/generate_agent_skill_docs.py b/scripts/generate_agent_skill_docs.py
index 29644ad04..8b4b62460 100644
--- a/scripts/generate_agent_skill_docs.py
+++ b/scripts/generate_agent_skill_docs.py
@@ -299,9 +299,11 @@ _OPERATION_NOTES: dict[str, list[str]] = {
     ],
     "media.recognize": [
         "- **Agent note:** Recognition returns a context containing `media_info`, not a candidate list. "
-        "Check its title and media type against the user's request. If recognition is missing or "
-        "ambiguous, use `media.search` and compare candidates by title, type, and year. Preserve "
-        "the selected result's exact `media_source` + `media_id` for `search.torrents`.",
+        "A failed match still returns parsed `meta_info` with the applied recognition words, so treat "
+        "recognition as successful only when `media_info` is present. Check its title and media type "
+        "against the user's request. If recognition is missing or ambiguous, use `media.search` and "
+        "compare candidates by title, type, and year. Preserve the selected result's exact "
+        "`media_source` + `media_id` for `search.torrents`.",
     ],
     "search.torrents": [
         "- **Agent note:** `media_id` is a **path** parameter and MUST be nested under "
diff --git a/skills/moviepilot-api/api/media.md b/skills/moviepilot-api/api/media.md
index daf7b9262..15457e9ed 100644
--- a/skills/moviepilot-api/api/media.md
+++ b/skills/moviepilot-api/api/media.md
@@ -182,7 +182,7 @@ Identity scope: `bound_user`; required role: `user`.
 - `path_params`: none
 - `query`: `custom_words` (string|null): Custom recognition or rename words applied to this media workflow.; `media_source` (MediaSource|null): Metadata source identifier. Preserve the exact value returned with media_id.; `subtitle` (string|null): Optional subtitle text used together with title during media recognition.; `title*` (string): Media, torrent, subscription, or history title used by the operation.
 - `body`: none
-- **Agent note:** Recognition returns a context containing `media_info`, not a candidate list. Check its title and media type against the user's request. If recognition is missing or ambiguous, use `media.search` and compare candidates by title, type, and year. Preserve the selected result's exact `media_source` + `media_id` for `search.torrents`.
+- **Agent note:** Recognition returns a context containing `media_info`, not a candidate list. A failed match still returns parsed `meta_info` with the applied recognition words, so treat recognition as successful only when `media_info` is present. Check its title and media type against the user's request. If recognition is missing or ambiguous, use `media.search` and compare candidates by title, type, and year. Preserve the selected result's exact `media_source` + `media_id` for `search.torrents`.
 
 ### `media.recognize_file`
 `GET /api/v1/media/recognize_file`; policy effect: `safe_read`.
diff --git a/tests/test_media_recognize_endpoint.py b/tests/test_media_recognize_endpoint.py
index c789f94a7..7d89f22cb 100644
--- a/tests/test_media_recognize_endpoint.py
+++ b/tests/test_media_recognize_endpoint.py
@@ -87,3 +87,22 @@ def test_recognize_audio_file_title_uses_music_meta_without_source() -> None:
     metainfo = chain.async_recognize_by_meta.await_args.args[0]
     assert isinstance(metainfo, MetaMusic)
     assert metainfo.audio_format
+
+
+def test_recognize_returns_meta_info_when_media_lookup_fails() -> None:
+    """未匹配到媒体时也应返回解析元数据，供识别测试展示识别词应用情况和解析结果。"""
+    chain = Mock()
+    chain.async_recognize_by_meta = AsyncMock(return_value=None)
+
+    with patch("app.api.endpoints.media.MediaChain", return_value=chain):
+        result = asyncio.run(
+            recognize(
+                title="Some.Release.2024",
+                custom_words="Some => Other",
+                _=Mock(),
+            )
+        )
+
+    assert result["media_info"] is None
+    assert result["meta_info"]["org_string"] == "Other.Release.2024"
+    assert result["meta_info"]["apply_words"] == ["Some => Other"]