Video by URL or local path has to be encoded and sent inline, which puts a hard ceiling on file size; the YouTube tool hands over a link instead and sidesteps that entirely. Knowing which of the five tools applies is most of what you need to use this well. The multi-item calls are the other real capability — comparing two screenshots or picking one image out of several is a different question from describing one, and the tools are shaped for it.
An MCP server that hands images and videos to Gemini 2.0 Flash for analysis. It accepts remote URLs, local file paths and YouTube links, takes several items in one call so relationships between them can be described, and lets you steer the analysis with an optional prompt.
- One or more images analysed from URLs, with an optional prompt — `analyze_image` taking `imageUrls`
- The same for local image files — `analyze_image_from_path` taking `imagePaths`
- Videos analysed from URLs or from local paths, best suited to smaller files — `analyze_video`, `analyze_video_from_path`
- A YouTube video analysed straight from its link, with no download step and none of the size limit that applies to the other video paths — `analyze_youtube_video`
- Multiple items in one call, so a question about the relationship between them — which of these contains a cat, what changed between these two screenshots — is answerable
- URL validity checking, local files loaded and base64-encoded, and basic security checks on the paths it is given
A Gemini API key, supplied as `GEMINI_API_KEY`. Install by cloning, `npm install` and `npm run build`, then pointing the client's command at node and `dist/index.js` with the key in the env block; a Smithery installer is also available. Two things to plan around: video sent by URL or path is capped by what Gemini accepts inline — around 20MB after base64 encoding — so larger files fail while YouTube analysis is unaffected; and supported video types are a specific list including `video/mp4`, `video/mpeg`, `video/webm` and `video/3gpp`, with others skipped. For the path-based tools, the assistant must give paths valid in the environment the server runs in, converting between Windows and WSL forms itself. The npm package name is `image-mcp-server` (1.0.0). Forked from champierre/image-mcp-server.
One command plus a key — npx -y @smithery/cli install @Rentapad/image-mcp-server --client claude, then supply credentials