Gemini MCP Server

# Gemini MCP Server - Complete Parameters Reference This guide provides detailed information on all available parameters for each tool. ## 1. generate_text Tool ### Required Parameters - **prompt** (string): The text prompt to send to Gemini ### Optional Parameters | Parameter | Type | Default | Description | Example Values | |-----------|------|---------|-------------|----------------| | **model** | string | gemini-2.5-flash | Gemini model to use | gemini-2.5-pro, gemini-2.5-flash, gemini-2.0-flash | | **systemInstruction** | string | none | System prompt to guide behavior | "You are a helpful Python tutor" | | **temperature** | number | 0.7 | Creativity level (0-2) | 0.1 (precise), 0.7 (balanced), 1.5 (creative) | | **maxTokens** | number | 2048 | Maximum output tokens | 100, 500, 1000, 4096 | | **topK** | number | 40 | Top-k sampling | 1 (greedy), 40 (default), 100 (diverse) | | **topP** | number | 0.95 | Nucleus sampling | 0.1 (focused), 0.95 (default), 1.0 (all) | | **jsonMode** | boolean | false | Enable JSON output | true, false | | **jsonSchema** | object | none | JSON schema validation | See examples below | | **grounding** | boolean | false | Enable Google Search | true, false | | **conversationId** | string | none | Maintain conversation | "chat-001", "session-123" | | **safetySettings** | array | default | Content filtering | See safety section | ### Examples with Parameters #### Basic Text Generation ``` "Use Gemini to explain machine learning" ``` #### With Specific Model and Temperature ``` "Use Gemini 2.5 Pro with temperature 0.2 to write technical documentation for this API" ``` #### With System Instruction ``` "Use Gemini with system instruction 'You are an expert Python developer' to review this code" ``` #### JSON Mode with Schema ``` "Use Gemini in JSON mode to analyze this text and return: { type: object, properties: { sentiment: { type: string, enum: ['positive', 'negative', 'neutral'] }, confidence: { type: number, minimum: 0, maximum: 1 }, keywords: { type: array, items: { type: string } } } }" ``` #### With Grounding ``` "Use Gemini with grounding enabled to tell me about the latest AI developments" ``` #### With Conversation Memory ``` "Start a conversation with ID 'python-help' and ask Gemini about decorators" "Continue conversation 'python-help' and ask about generators" ``` #### With Safety Settings ``` "Use Gemini with safety settings [ { category: 'HARM_CATEGORY_HARASSMENT', threshold: 'BLOCK_NONE' } ] to analyze this medical text" ``` ## 2. analyze_image Tool ### Required Parameters (one of these) - **imageUrl** (string): URL of the image to analyze - **imageBase64** (string): Base64-encoded image data ### Required Parameters - **prompt** (string): Question or instruction about the image ### Optional Parameters | Parameter | Type | Default | Description | Example Values | |-----------|------|---------|-------------|----------------| | **model** | string | gemini-2.5-flash | Vision-capable model | gemini-2.5-pro, gemini-2.5-flash, gemini-2.0-flash | ### Examples #### With Image URL ``` "Analyze this image and describe what you see: https://example.com/image.jpg" ``` #### With Base64 Image ``` "What's in this screenshot? [paste image directly]" ``` #### With Specific Model ``` "Use Gemini 2.5 Pro to analyze this architecture diagram and explain the components" ``` ## 3. count_tokens Tool ### Required Parameters - **text** (string): Text to count tokens for ### Optional Parameters | Parameter | Type | Default | Description | Example Values | |-----------|------|---------|-------------|----------------| | **model** | string | gemini-2.5-flash | Model for token counting | Any Gemini model | ### Examples ``` "Count tokens for this text: [your long text]" "How many tokens would this use with gemini-2.5-pro: [text]" ``` ## 4. list_models Tool ### Optional Parameters | Parameter | Type | Default | Description | Example Values | |-----------|------|---------|-------------|----------------| | **filter** | string | all | Filter by capability | all, thinking, vision, grounding, json_mode | ### Examples ``` "List all Gemini models" "Show me models with thinking capability" "Which models support grounding?" "List models that have JSON mode" ``` ## 5. embed_text Tool ### Required Parameters - **text** (string): Text to generate embeddings for ### Optional Parameters | Parameter | Type | Default | Description | Example Values | |-----------|------|---------|-------------|----------------| | **model** | string | text-embedding-004 | Embedding model | text-embedding-004, text-multilingual-embedding-002 | ### Examples ``` "Generate embeddings for: Machine learning is fascinating" "Create multilingual embeddings for this text using text-multilingual-embedding-002" ``` ## Advanced Parameter Combinations ### Complex Analysis with All Features ``` "Use Gemini 2.5 Pro to analyze this code with: - System instruction: 'You are a security expert' - Temperature: 0.3 - Max tokens: 4096 - JSON mode enabled - Schema: { security_score: number, vulnerabilities: array, recommendations: array } - Grounding enabled for latest security practices" ``` ### Creative Writing with Parameters ``` "Use Gemini 2.5 Flash to write a story with: - Temperature: 1.5 - Max tokens: 2000 - Top-k: 100 - Top-p: 0.98 - System instruction: 'You are a creative sci-fi writer'" ``` ### Conversation with Context ``` "Start conversation 'code-review-001' with Gemini 2.5 Pro using: - System instruction: 'You are a thorough code reviewer' - Temperature: 0.5 - Review this Python function" "Continue conversation 'code-review-001' and ask about performance optimizations" ``` ## Safety Settings Reference ### Categories - HARM_CATEGORY_HARASSMENT - HARM_CATEGORY_HATE_SPEECH - HARM_CATEGORY_SEXUALLY_EXPLICIT - HARM_CATEGORY_DANGEROUS_CONTENT ### Thresholds - BLOCK_NONE - No blocking - BLOCK_ONLY_HIGH - Block only high probability - BLOCK_MEDIUM_AND_ABOVE - Block medium and high - BLOCK_LOW_AND_ABOVE - Block all but negligible ### Example ``` "Use Gemini with safety settings: [ { category: 'HARM_CATEGORY_HARASSMENT', threshold: 'BLOCK_ONLY_HIGH' }, { category: 'HARM_CATEGORY_DANGEROUS_CONTENT', threshold: 'BLOCK_NONE' } ] to analyze this medical research paper" ``` ## Model-Specific Features ### Thinking Models (2.5 series) - gemini-2.5-pro - gemini-2.5-flash - gemini-2.5-flash-lite These models have enhanced reasoning capabilities. Use them for: - Complex problem solving - Code analysis and generation - Deep explanations - Multi-step reasoning ### Models with Grounding - gemini-2.5-pro - gemini-2.5-flash - gemini-2.0-flash - gemini-2.0-pro-experimental These can access Google Search for current information. ### All Models Support - JSON mode - System instructions - Function calling - Safety settings - Temperature control ## Tips for Parameter Usage 1. **Start Simple**: Begin with just the prompt, add parameters as needed 2. **Model Selection**: Use 2.5-flash for most tasks, 2.5-pro for complex reasoning 3. **Temperature**: Lower for factual tasks, higher for creative tasks 4. **Token Limits**: Count tokens first for long inputs 5. **JSON Mode**: Always provide a schema for consistent output 6. **Grounding**: Enable only when you need current information 7. **Conversations**: Use IDs to maintain context across multiple queries ## Common Parameter Patterns ### For Code Review ``` model: "gemini-2.5-pro" temperature: 0.3 systemInstruction: "You are an expert code reviewer" jsonMode: true maxTokens: 4096 ``` ### For Creative Writing ``` model: "gemini-2.5-flash" temperature: 1.2 topK: 100 topP: 0.98 maxTokens: 2000 ``` ### For Factual Analysis ``` model: "gemini-2.5-flash" temperature: 0.1 grounding: true jsonMode: true ``` ### For Learning/Tutoring ``` model: "gemini-2.5-flash" systemInstruction: "You are a patient teacher" temperature: 0.7 conversationId: "learning-session-001" ```