Loading ToolBench
MCP server bringing Gemini ingestion (audio/video/document digest), image & video generation, and Vertex AI Search to Claude Desktop.
Server has 8 tools with mostly complete schemas and descriptions. Naming follows verb_noun convention (digest, generate_image, generate_video, search_internal, list_search_engines, health_check, auth_setup, savings_report). Descriptions are present and detailed (194-500+ chars), exceeding baseline. However, several tools lack parameter descriptions (e.g., generate_video 'quality' enum lacks explanation of what 'fast' vs 'standard' means), and output schemas are not documented. Error handling is present but generic. Tool composition is reasonable but some tools conflate concerns (e.g., digest handles both documents and media).
Set up authentication for the MCP server
Read and summarise files with Gemini's long context: PDFs, documents, slides, images, audio and video. Accepts local paths, Google Drive links and gs:// Cloud Storage URIs, mixed freely in one call. Returns a compact digest with citations (page/section for documents, timestamps for recordings), or answers a specific question about the files. Handles very large files and multiple files at once, and detects the file type automatically. Always prefer this over reading a large file into the conversation yourself: the bulk stays in Gemini and only the digest comes back. 資料・PDF・ドキュメント・スライドを読んで要約する。会議の録画・録音・動画・音声の文字起こしや議事録、要点・決定事項・アクションアイテムの抽出。動画の画面に何が映っているかの説明。長い資料をまとめて読んで要点だけ知りたいとき。
Generate images using Gemini's image generation capability
Generate videos using Gemini's video generation capability with progress notifications
Check the health and authentication status of the MCP server
Output schemas not documented. Tools return structured data (usage blocks, images, videos, search results) but LLMs cannot see what fields to expect. This forces agents to guess at response structure and risks extraction errors.
Parameter descriptions incomplete or missing. generate_video 'quality' enum (fast|standard) lacks explanation of tradeoffs. list_search_engines and health_check have empty input schemas but no description of what they return or when to call them.
digest tool conflates two concerns: document summarization and media transcription. While the implementation may handle both, splitting into digest_document and digest_media would clarify intent and let agents compose them independently.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | <=2025-11-25 | v2 |
List available Vertex AI Search engines
Generate a report of token savings and cost avoidance from using this MCP server
Search using Vertex AI Search with a specified search engine
Error handling is generic. Code shows fail() returns plain text warnings without categorization (retryable vs user-fixable vs fatal). Agents cannot determine whether to retry, ask the user, or abort.
No dry-run or confirmation step for destructive/generative tools. generate_image and generate_video create assets without confirmation. Agents may accidentally generate unwanted content.