Provides tools for interacting with Google's Gemini API, enabling chat, multimodal analysis, deep research, file management, and various generation tasks.
Allows analysis of public YouTube videos by URL, extracting content for processing via Gemini's multimodal capabilities.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Gemini MCP Server analyze this YouTube video: https://youtube.com/watch?v=dQw4w9WgXcQ"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A Model Context Protocol (MCP) server providing 23 tools for Google's Gemini API -- chat, multimodal analysis, deep research, file management, YouTube analysis, and more.
Built with @google/genai SDK (v1.0.0+).
Chat with Gemini models (single-turn, multi-turn, with tool modes)
Analyze images, audio, video, PDFs, YouTube videos, and URLs
Files API with auto-switching: inline for small files (<20MB), upload for large (up to 2GB)
Deep research agent with background polling and push notifications (Termux)
Structured JSON output, embeddings, code execution, translation, summarization
Google Search grounding and URL context
Thinking mode enabled by default (budget: 65535 tokens)
High media resolution by default
Related MCP server: Gemini Chat MCP
Node.js 18+
git clone https://github.com/salviz/gemini-mcp-server.git
cd gemini-mcp-server
npm installCLI:
claude mcp add gemini -- node /path/to/gemini-mcp-server/index.jsOr add to your MCP config (~/.claude/claude_desktop_config.json or .mcp.json):
{
"mcpServers": {
"gemini": {
"type": "stdio",
"command": "node",
"args": ["/path/to/gemini-mcp-server/index.js"],
"env": {
"GEMINI_API_KEY": "your-api-key"
}
}
}
}Variable | Required | Description |
| Yes | Your Google Gemini API key |
Tool | Description |
| Send a prompt with optional search grounding and URL context |
| Multi-turn conversation with message history |
| Chat with mode switching: |
| Search-grounded generation with source citations |
| Generate JSON output matching a provided schema |
| Analyze one or more URLs using Gemini's URL context tool |
Tool | Description |
| Analyze an image file with Gemini Vision (JPG, PNG, GIF, WebP, BMP, SVG) |
| Transcribe, summarize, or describe audio (MP3, WAV, OGG, FLAC, AAC, M4A, Opus) |
| Analyze a video file; auto-uploads large files via Files API (MP4, AVI, MOV, MKV, WebM) |
| Analyze a PDF document (up to 2GB via Files API) |
| Analyze a public YouTube video by URL (no download needed) |
| Analyze content from an HTTP/HTTPS URL or GCS URI ( |
Tool | Description |
| Start a deep research task; sends push notification on completion |
| Check status of a running deep research task by interaction ID |
Tool | Description |
| Upload a file to Gemini (up to 2GB, retained 48 hours) |
| List all uploaded files with metadata |
| Delete an uploaded file by name |
Tool | Description |
| List available Gemini models with capabilities and token limits |
| Count tokens in text using a model's tokenizer |
| Generate text embeddings (default: |
| Execute Python code via Gemini's built-in sandbox |
| Summarize text with configurable style (brief, detailed, bullet-points) |
| Translate text to any language with optional model override |
Default model: gemini-3.1-pro-preview. Every tool accepts an optional model parameter.
Model | Best For |
| Default. Best quality for most tasks |
| Faster responses, lower cost |
| Text embeddings (used by |
| Deep research agent (used internally) |
The server automatically handles file size:
<= 20MB: Sent inline as base64 (fast, no upload step)
> 20MB up to 2GB: Uploaded via Gemini Files API, then referenced by URI
YouTube URLs: Passed directly via fileData.fileUri (no download)
HTTP/HTTPS URLs: Passed via createPartFromUri (up to 100MB)
GCS URIs (gs://): Passed via fileData.fileUri
Uploaded files are retained for 48 hours. Use gemini_list_files and gemini_delete_file to manage them.
The gemini_deep_research tool uses Gemini's Interactions API with the deep-research-pro-preview-12-2025 agent:
Starts research in background mode
Polls for 50 seconds in case it finishes quickly
If still running, starts background polling (every 30s, up to 30 minutes)
Sends a push notification via termux-notification when complete
Saves full results to ~/.cache/deep_research_*.txt
Use gemini_check_research to manually poll at any time.
gemini-mcp-server/
index.js # Server entry point
tools/
shared.js # Shared config, AI client, extractText helper
chat.js # 17 tools: chat, analysis, research, files, YouTube
utility.js # 6 tools: models, tokens, embed, code, summarize, translate
package.jsonAPI key from environment only -- never hardcoded in source
File paths validated -- absolute paths required, existence checked before reading
Stdio transport -- no network server exposed
No data logged or stored -- prompts and responses are not persisted
Package | Version | Purpose |
| ^1.0.0 | MCP server framework |
| ^1.0.0 | Google Gemini AI SDK |
| ^3.24.0 | Input schema validation |
MIT
Analyze an audio file with Gemini (transcribe, summarize, describe)
Inputs: audioPath (required), model, prompt.
Analyze an image file with Gemini Vision
Inputs: imagePath (required), model, prompt.
Analyze a PDF document with Gemini (summarize, extract, answer questions)
Inputs: pdfPath (required), model, prompt.
Analyze a file from a public URL (HTTP/HTTPS) or Google Cloud Storage URI (gs://). Supports images, audio, video, PDFs, and documents up to 100MB.
Inputs: url (required), mimeType (required), model, prompt.
Analyze a video file with Gemini (describe, summarize, extract info)
Inputs: videoPath (required), model, prompt.
Analyze a public YouTube video with Gemini (summarize, transcribe, answer questions). Supports up to 10 videos with Gemini 2.5+ models.
Inputs: url (required), model, prompt.
Send a prompt to Gemini and get a text response. Optionally enable search grounding or URL context.
Inputs: prompt (required), model, temperature, enableSearch, +3 more.
Multi-turn conversation with Gemini using message history
Inputs: messages (required), model, systemInstruction.
Flexible Gemini chat with mode switching: "search" (Google Search + URL context), "code" (code execution), or "all" (everything)
Inputs: prompt (required), mode, model, systemInstruction.
Check or wait for deep research results. Use wait=true to block until completion (ideal for background execution — the tool returns when research finishes, triggering a task notification in your AI client).
Inputs: interactionId (required), wait, timeoutMinutes.
Generate and execute code using Gemini's built-in code execution capability
Inputs: prompt (required), model.
Count the number of tokens in the provided text using a Gemini model
Inputs: text (required), model.
Start a deep research task using Gemini Deep Research. Returns an interaction ID. Then call gemini_check_research with wait=true in the background to get notified when complete.
Inputs: query (required).
Delete a file from the Gemini Files API
Inputs: fileName (required).
Generate text embeddings using a Gemini embedding model
Inputs: text (required), model.
List files uploaded to the Gemini Files API
Inputs: pageSize.
List available Gemini models with their capabilities and token limits
Inputs: no named parameters.
Send a prompt to Gemini with Google Search grounding and URL context for up-to-date information
Inputs: prompt (required), model, enableUrlContext.
Get structured JSON output from Gemini using a provided JSON schema
Inputs: prompt (required), jsonSchema (required), model.
Summarize long text using Gemini with configurable summary style
Inputs: text (required), model, style.
Translate text to a target language using Gemini
Inputs: text (required), targetLanguage (required), model, sourceLanguage.
Upload a file to Gemini Files API for use with analysis tools. Files are retained for 48 hours. Max 2GB per file, 20GB total.
Inputs: filePath (required), mimeType, displayName.
Analyze one or more URLs using Gemini URL context tool. Gemini fetches and reads the URLs directly.
Inputs: prompt (required), urls, model.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 23 tools
Most tools have distinct purposes, but there is some overlap among chat variants (e.g., gemini_chat, gemini_chat_multi, gemini_chat_with_tools) and analysis tools (e.g., analyze_pdf vs. summarize). Descriptions help, but an agent might occasionally select the wrong tool.
All tools follow the consistent pattern 'gemini_verb_noun' (e.g., gemini_analyze_audio, gemini_list_files). The naming is uniform and predictable, making it easy for an agent to infer tool purposes.
23 tools cover a broad range of capabilities. While the count is slightly high, each tool serves a distinct function within the Gemini ecosystem. A few similarities exist, but overall the scope is justified.
The tool set covers all major Gemini features: multimodal analysis, chat variants, file management, embeddings, code execution, search, structured output, and more. There are no obvious gaps for typical use cases.
YouTube transcripts, search, channels, playlists and bulk transcript jobs for AI agents. 14 tools.
37 tools: LLM chat, summaries, translation, embeddings, speech, images. 25 free calls.
23 live research tools, from web search to SEC filings. Each call bills the scraper it runs.
Web scraping, Google SERP, Markdown and ChatGPT/Gemini answers as typed tools for AI agents.