Routes media, document, internal-search and image/video-generation requests to the gemini-studio-mcp server. Use when the user wants a meeting recording, video or audio summarised or transcribed; a PDF, deck or long document read, digested or asked questions about; company-internal information looked up; or an image or video generated. 会議の録画・録音の要約や議事録、動画の内容説明、PDF・資料・スライドを読んで要約、社内ドキュメント・規程・手順書の検索、画像やキービジュアルの生成、動画生成。
Installs into .claude/skills of the current project.
Are you the author of Gemini Studio?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/yuting0624-gemini-studio)
---
name: gemini-studio
description: Routes media, document, internal-search and image/video-generation requests to the gemini-studio-mcp server. Use when the user wants a meeting recording, video or audio summarised or transcribed; a PDF, deck or long document read, digested or asked questions about; company-internal information looked up; or an image or video generated. 会議の録画・録音の要約や議事録、動画の内容説明、PDF・資料・スライドを読んで要約、社内ドキュメント・規程・手順書の検索、画像やキービジュアルの生成、動画生成。
---
# Gemini Studio routing
Delegate to the `gemini-studio-mcp` server instead of reading large files or
searching the web. Gemini handles the intake; you do the reasoning on the digest.
## Which tool
| Request | Tool |
|---|---|
| Recording, video, audio, meeting, 議事録, 文字起こし | `gemini-studio-mcp:digest` |
| PDF, deck, doc, spreadsheet-as-PDF, image, 資料, 要約 | `gemini-studio-mcp:digest` |
| Anything the company knows but the public web does not | `gemini-studio-mcp:search_internal` |
| Which internal data exists | `gemini-studio-mcp:list_search_engines` |
| Make an image, キービジュアル, バナー, モックアップ | `gemini-studio-mcp:generate_image` |
| Make a video, ショート動画 | `gemini-studio-mcp:generate_video` |
`digest` takes documents *and* recordings and routes on file type. Do not try to
pick a media-specific or document-specific tool; there is only `digest`.
## Rules
**Pass filenames through as given.** `digest` resolves a bare name against
Downloads, Desktop and Documents, and reports the candidates when several match.
Do not guess a directory or ask the user for a full path first.
**Do not open the file yourself.** The point is that the bytes never enter this
conversation. Read the digest, not the source.
**Default to flash.** Only pass `tier: "pro"` when the user needs precise
timestamps, speaker separation, or hard reasoning over the material — it costs
several times more and exhausts quota that later calls need.
**Answer language follows the user.** Omit `language` and the digest comes back
in the language the question was asked in. Only set it when the user asks for a
specific language.
**Search internal data before answering from general knowledge.** For questions
about procedures, policies, products or anything organisation-specific, try
`search_internal` first. Omit `engine_id` unless the user names a source; the
server picks the default or tells you the options.
**Generation is chargeable and leaves files.** Confirm intent before generating
more than one image, or any video. Report the saved path back to the user —
images and videos are written to `~/GeminiMCP/outputs/` and only a preview is
shown inline.
## When something fails
Every error from this server already states what to do next; relay it rather
than reinterpreting it. Only call `gemini-studio-mcp:health_check` when a call
has actually failed, or when the user asks about setup — it is a diagnostic, not
a warm-up. `gemini-studio-mcp:auth_setup` opens a browser, so never call it
without the user asking.
If a call fails once with an authentication error, just retry: the server
refreshes credentials itself.