# How to integrate Gemini MCP with Vercel AI SDK v6 ```json { "title": "How to integrate Gemini MCP with Vercel AI SDK v6", "toolkit": "Gemini", "toolkit_slug": "gemini", "framework": "Vercel AI SDK", "framework_slug": "ai-sdk", "url": "https://composio.dev/toolkits/gemini/framework/ai-sdk", "markdown_url": "https://composio.dev/toolkits/gemini/framework/ai-sdk.md", "updated_at": "2026-05-12T10:12:37.656Z" } ``` ## Introduction This guide walks you through connecting Gemini to Vercel AI SDK v6 using the Composio tool router. By the end, you'll have a working Gemini agent that can summarize this research article in 100 words, generate a creative image of a futuristic city, create a 30-second video based on this script through natural language commands. This guide will help you understand how to give your Vercel AI SDK agent real control over a Gemini account through Composio's Gemini MCP server. Before we dive in, let's take a quick look at the key ideas and tools involved. ## Also integrate Gemini with - [ChatGPT](https://composio.dev/toolkits/gemini/framework/chatgpt) - [OpenAI Agents SDK](https://composio.dev/toolkits/gemini/framework/open-ai-agents-sdk) - [Claude Agent SDK](https://composio.dev/toolkits/gemini/framework/claude-agents-sdk) - [Claude Code](https://composio.dev/toolkits/gemini/framework/claude-code) - [Claude Cowork](https://composio.dev/toolkits/gemini/framework/claude-cowork) - [Codex](https://composio.dev/toolkits/gemini/framework/codex) - [Cursor](https://composio.dev/toolkits/gemini/framework/cursor) - [VS Code](https://composio.dev/toolkits/gemini/framework/vscode) - [OpenCode](https://composio.dev/toolkits/gemini/framework/opencode) - [OpenClaw](https://composio.dev/toolkits/gemini/framework/openclaw) - [Hermes](https://composio.dev/toolkits/gemini/framework/hermes-agent) - [CLI](https://composio.dev/toolkits/gemini/framework/cli) - [Google ADK](https://composio.dev/toolkits/gemini/framework/google-adk) - [LangChain](https://composio.dev/toolkits/gemini/framework/langchain) - [Mastra AI](https://composio.dev/toolkits/gemini/framework/mastra-ai) - [LlamaIndex](https://composio.dev/toolkits/gemini/framework/llama-index) - [CrewAI](https://composio.dev/toolkits/gemini/framework/crew-ai) ## TL;DR Here's what you'll learn: - How to set up and configure a Vercel AI SDK agent with Gemini integration - Using Composio's Tool Router to dynamically load and access Gemini tools - Creating an MCP client connection using HTTP transport - Building an interactive CLI chat interface with conversation history management - Handling tool calls and results within the Vercel AI SDK framework ## What is Vercel AI SDK? The Vercel AI SDK is a TypeScript library for building AI-powered applications. It provides tools for creating agents that can use external services and maintain conversation state. Key features include: - streamText: Core function for streaming responses with real-time tool support - MCP Client: Built-in support for Model Context Protocol via @ai-sdk/mcp - Step Counting: Control multi-step tool execution with stopWhen: stepCountIs() - OpenAI Provider: Native integration with OpenAI models ## What is the Gemini MCP server, and what's possible with it? The Gemini MCP server is an implementation of the Model Context Protocol that connects your AI agent and assistants like Claude, Cursor, etc directly to your Gemini account. It provides structured and secure access to Gemini's multimodal AI features, so your agent can generate text, images, and videos, analyze content, and manage model resources on your behalf. - Text and content generation: Instruct your agent to create high-quality, customized text using Gemini's advanced generative models—great for brainstorming, drafting, or summarizing information. - Creative image and video generation: Ask the agent to generate original images or high-quality videos from text prompts using Gemini 2.5 Flash and Veo models, with fine control over style and format. - Embedding and semantic analysis: Let your agent transform any text into rich semantic embeddings for similarity search, clustering, or classification tasks. - Model discovery and optimization: Have the agent list available Gemini and Veo models, check their capabilities, and select the best fit for your project or workflow. - Efficient resource management: Enable the agent to track video generation operations, download final assets, and optimize prompt inputs by counting tokens—all without manual intervention. ## Supported Tools | Tool slug | Name | Description | |---|---|---| | `GEMINI_COUNT_TOKENS` | Count Tokens (Gemini) | Counts the number of tokens in text using Gemini tokenization. Useful for estimating costs, checking input limits, and optimizing prompts before making API calls. | | `GEMINI_EMBED_CONTENT` | Embed Content (Gemini) | Generates text embeddings using Gemini embedding models. Converts text into numerical vectors for semantic search, similarity comparison, clustering, and classification tasks. | | `GEMINI_GENERATE_CONTENT` | Generate Content (Gemini) | Generates text content or speech audio from prompts using Gemini models. Supports text generation models (Gemini Flash, Pro) and text-to-speech models with configurable parameters. Generated text is nested at results[i].response.data.text. Output may be wrapped in markdown fences (e.g., ```html...```) or preceded by explanatory prose; strip these before file writing or rendering. | | `GEMINI_GENERATE_IMAGE` | Generate Image (Nano Banana) | Generates images from text prompts using Gemini models (Nano Banana). Supports models: 'gemini-2.5-flash-image' (GA stable, fast), 'gemini-3-pro-image-preview' (Nano Banana Pro - advanced with 4K resolution, thinking mode, up to 14 reference images), and 'gemini-2.0-flash-exp-image-generation' (2.0 Flash experimental). Returns one image per call; images are uploaded to S3. Parse response at data.image.s3url or the text-type entry in data.content — prefer the URL to avoid base64 blobs. Always validate s3url before treating call as successful; a 200 response may contain only text with no image. Store s3url immediately as URLs can expire. Output formats are raster only (JPG/PNG/WebP); request PNG for transparency. Concurrent usage may trigger HTTP 429/RESOURCE_EXHAUSTED — keep concurrency ≤3 and use exponential backoff (1s→2s→4s, ~5 retries). NOTE NEVER EVER TRUE SYNC_TO_WORKBENCH IN RUBE_MULTI_EXECUTE_TOOL | | `GEMINI_GENERATE_VIDEOS` | Generate Videos (Veo) | Generates videos from text prompts using Google's Veo models. Returns an operation_name for tracking; pass it verbatim (no edits) to GEMINI_WAIT_FOR_VIDEO or GEMINI_GET_VIDEOS_OPERATION. Jobs take 30–180+ seconds; wait 10s before first poll, then poll every 10–30s (allow up to 12 min). Successful results include data.video_file.s3url — missing s3url means failure. If done=true but no video_file, check raiMediaFilteredReasons (safety block); revise prompt and regenerate. Text-only; cannot accept image inputs. Max ~3–5 concurrent jobs; 429 RESOURCE_EXHAUSTED requires exponential backoff. For retries, always start a fresh call — never reuse a failed operation_name. | | `GEMINI_LIST_MODELS` | List Models (Gemini API) | Lists available Gemini and Veo models with their capabilities and limits. Useful for discovering supported models and their features before making generation requests. Before calling video generation tools, verify model availability here — preview Veo models (e.g., veo-3.0-generate-preview) may be unavailable or return missing video URIs; prefer stable models like veo-2.0-generate-001. | | `GEMINI_WAIT_FOR_VIDEO` | Wait and Download Video (Veo) | Polls a Veo video generation operation until completion, then downloads and returns the video as a FileDownloadable. Generation takes 30–120+ seconds (up to ~10–12 min); long waits are normal, not failures. On completion, the URL is nested at data.video_file.s3url — validate it is non-empty before downstream use. A done=true response without a valid s3url indicates safety filter rejection (check raiMediaFilteredReasons) or quota exhaustion — adjust the prompt and regenerate. On timeout, use GEMINI_GET_VIDEOS_OPERATION with incremental backoff before starting a new job. Keep parallel jobs to 3–5 to avoid 429 RESOURCE_EXHAUSTED errors. | ## Supported Triggers None listed. ## Creating MCP Server - Stand-alone vs Composio SDK The Gemini MCP server is an implementation of the Model Context Protocol that connects your AI agent to Gemini. It provides structured and secure access so your agent can perform Gemini operations on your behalf through a secure, permission-based interface. With Composio's managed implementation, you don't have to create your own developer app. For production, if you're building an end product, we recommend using your own credentials. The managed server helps you prototype fast and go from 0-1 faster. ## Step-by-step Guide ### 1. Prerequisites Before you begin, make sure you have: - Node.js and npm installed - A Composio account with API key - An OpenAI API key ### 1. Getting API Keys for OpenAI and Composio OpenAI API Key - Go to the [OpenAI dashboard](https://platform.openai.com/settings/organization/api-keys) and create an API key. You'll need credits to use the models, or you can connect to another model provider. - Keep the API key safe. Composio API Key - Log in to the [Composio dashboard](https://dashboard.composio.dev?utm_source=toolkits&utm_medium=framework_docs). - Navigate to your API settings and generate a new API key. - Store this key securely as you'll need it for authentication. ### 2. Install required dependencies First, install the necessary packages for your project. What you're installing: - @ai-sdk/openai: Vercel AI SDK's OpenAI provider - @ai-sdk/mcp: MCP client for Vercel AI SDK - @composio/core: Composio SDK for tool integration - ai: Core Vercel AI SDK - dotenv: Environment variable management ```bash npm install @ai-sdk/openai @ai-sdk/mcp @composio/core ai dotenv ``` ### 3. Set up environment variables Create a .env file in your project root. What's needed: - OPENAI_API_KEY: Your OpenAI API key for GPT model access - COMPOSIO_API_KEY: Your Composio API key for tool access - COMPOSIO_USER_ID: A unique identifier for the user session ```bash OPENAI_API_KEY=your_openai_api_key_here COMPOSIO_API_KEY=your_composio_api_key_here COMPOSIO_USER_ID=your_user_id_here ``` ### 4. Import required modules and validate environment What's happening: - We're importing all necessary libraries including Vercel AI SDK's OpenAI provider and Composio - The dotenv/config import automatically loads environment variables - The MCP client import enables connection to Composio's tool server ```typescript import "dotenv/config"; import { openai } from "@ai-sdk/openai"; import { Composio } from "@composio/core"; import * as readline from "readline"; import { streamText, type ModelMessage, stepCountIs } from "ai"; import { createMCPClient } from "@ai-sdk/mcp"; const composioAPIKey = process.env.COMPOSIO_API_KEY; const composioUserID = process.env.COMPOSIO_USER_ID; if (!process.env.OPENAI_API_KEY) throw new Error("OPENAI_API_KEY is not set"); if (!composioAPIKey) throw new Error("COMPOSIO_API_KEY is not set"); if (!composioUserID) throw new Error("COMPOSIO_USER_ID is not set"); const composio = new Composio({ apiKey: composioAPIKey, }); ``` ### 5. Create Tool Router session and initialize MCP client What's happening: - We're creating a Tool Router session that gives your agent access to Gemini tools - The create method takes the user ID and specifies which toolkits should be available - The returned mcp object contains the URL and authentication headers needed to connect to the MCP server - This session provides access to all Gemini-related tools through the MCP protocol ```typescript async function main() { // Create a tool router session for the user const session = await composio.create(composioUserID!, { toolkits: ["gemini"], }); const mcpUrl = session.mcp.url; ``` ### 6. Connect to MCP server and retrieve tools What's happening: - We're creating an MCP client that connects to our Composio Tool Router session via HTTP - The mcp.url provides the endpoint, and mcp.headers contains authentication credentials - The type: "http" is important - Composio requires HTTP transport - tools() retrieves all available Gemini tools that the agent can use ```typescript const mcpClient = await createMCPClient({ transport: { type: "http", url: mcpUrl, headers: session.mcp.headers, // Authentication headers for the Composio MCP server }, }); const tools = await mcpClient.tools(); ``` ### 7. Initialize conversation and CLI interface What's happening: - We initialize an empty messages array to maintain conversation history - A readline interface is created to accept user input from the command line - Instructions are displayed to guide the user on how to interact with the agent ```typescript let messages: ModelMessage[] = []; console.log("Chat started! Type 'exit' or 'quit' to end the conversation.\n"); console.log( "Ask any questions related to gemini, like summarize my last 5 emails, send an email, etc... :)))\n", ); const rl = readline.createInterface({ input: process.stdin, output: process.stdout, prompt: "> ", }); rl.prompt(); ``` ### 8. Handle user input and stream responses with real-time tool feedback What's happening: - We use streamText instead of generateText to stream responses in real-time - toolChoice: "auto" allows the model to decide when to use Gemini tools - stopWhen: stepCountIs(10) allows up to 10 steps for complex multi-tool operations - onStepFinish callback displays which tools are being used in real-time - We iterate through the text stream to create a typewriter effect as the agent responds - The complete response is added to conversation history to maintain context - Errors are caught and displayed with helpful retry suggestions ```typescript rl.on("line", async (userInput: string) => { const trimmedInput = userInput.trim(); if (["exit", "quit", "bye"].includes(trimmedInput.toLowerCase())) { console.log("\nGoodbye!"); rl.close(); process.exit(0); } if (!trimmedInput) { rl.prompt(); return; } messages.push({ role: "user", content: trimmedInput }); console.log("\nAgent is thinking...\n"); try { const stream = streamText({ model: openai("gpt-5"), messages, tools, toolChoice: "auto", stopWhen: stepCountIs(10), onStepFinish: (step) => { for (const toolCall of step.toolCalls) { console.log(`[Using tool: ${toolCall.toolName}]`); } if (step.toolCalls.length > 0) { console.log(""); // Add space after tool calls } }, }); for await (const chunk of stream.textStream) { process.stdout.write(chunk); } console.log("\n\n---\n"); // Get final result for message history const response = await stream.response; if (response?.messages?.length) { messages.push(...response.messages); } } catch (error) { console.error("\nAn error occurred while talking to the agent:"); console.error(error); console.log( "\nYou can try again or restart the app if it keeps happening.\n", ); } finally { rl.prompt(); } }); rl.on("close", async () => { await mcpClient.close(); console.log("\n👋 Session ended."); process.exit(0); }); } main().catch((err) => { console.error("Fatal error:", err); process.exit(1); }); ``` ## Complete Code ```typescript import "dotenv/config"; import { openai } from "@ai-sdk/openai"; import { Composio } from "@composio/core"; import * as readline from "readline"; import { streamText, type ModelMessage, stepCountIs } from "ai"; import { createMCPClient } from "@ai-sdk/mcp"; const composioAPIKey = process.env.COMPOSIO_API_KEY; const composioUserID = process.env.COMPOSIO_USER_ID; if (!process.env.OPENAI_API_KEY) throw new Error("OPENAI_API_KEY is not set"); if (!composioAPIKey) throw new Error("COMPOSIO_API_KEY is not set"); if (!composioUserID) throw new Error("COMPOSIO_USER_ID is not set"); const composio = new Composio({ apiKey: composioAPIKey, }); async function main() { // Create a tool router session for the user const session = await composio.create(composioUserID!, { toolkits: ["gemini"], }); const mcpUrl = session.mcp.url; const mcpClient = await createMCPClient({ transport: { type: "http", url: mcpUrl, headers: session.mcp.headers, // Authentication headers for the Composio MCP server }, }); const tools = await mcpClient.tools(); let messages: ModelMessage[] = []; console.log("Chat started! Type 'exit' or 'quit' to end the conversation.\n"); console.log( "Ask any questions related to gemini, like summarize my last 5 emails, send an email, etc... :)))\n", ); const rl = readline.createInterface({ input: process.stdin, output: process.stdout, prompt: "> ", }); rl.prompt(); rl.on("line", async (userInput: string) => { const trimmedInput = userInput.trim(); if (["exit", "quit", "bye"].includes(trimmedInput.toLowerCase())) { console.log("\nGoodbye!"); rl.close(); process.exit(0); } if (!trimmedInput) { rl.prompt(); return; } messages.push({ role: "user", content: trimmedInput }); console.log("\nAgent is thinking...\n"); try { const stream = streamText({ model: openai("gpt-5"), messages, tools, toolChoice: "auto", stopWhen: stepCountIs(10), onStepFinish: (step) => { for (const toolCall of step.toolCalls) { console.log(`[Using tool: ${toolCall.toolName}]`); } if (step.toolCalls.length > 0) { console.log(""); // Add space after tool calls } }, }); for await (const chunk of stream.textStream) { process.stdout.write(chunk); } console.log("\n\n---\n"); // Get final result for message history const response = await stream.response; if (response?.messages?.length) { messages.push(...response.messages); } } catch (error) { console.error("\nAn error occurred while talking to the agent:"); console.error(error); console.log( "\nYou can try again or restart the app if it keeps happening.\n", ); } finally { rl.prompt(); } }); rl.on("close", async () => { await mcpClient.close(); console.log("\n👋 Session ended."); process.exit(0); }); } main().catch((err) => { console.error("Fatal error:", err); process.exit(1); }); ``` ## Conclusion You've successfully built a Gemini agent using the Vercel AI SDK with streaming capabilities! This implementation provides a powerful foundation for building AI applications with natural language interfaces and real-time feedback. Key features of this implementation: - Real-time streaming responses for a better user experience with typewriter effect - Live tool execution feedback showing which tools are being used as the agent works - Dynamic tool loading through Composio's Tool Router with secure authentication - Multi-step tool execution with configurable step limits (up to 10 steps) - Comprehensive error handling for robust agent execution - Conversation history maintenance for context-aware responses You can extend this further by adding custom error handling, implementing specific business logic, or integrating additional Composio toolkits to create multi-app workflows. ## How to build Gemini MCP Agent with another framework - [ChatGPT](https://composio.dev/toolkits/gemini/framework/chatgpt) - [OpenAI Agents SDK](https://composio.dev/toolkits/gemini/framework/open-ai-agents-sdk) - [Claude Agent SDK](https://composio.dev/toolkits/gemini/framework/claude-agents-sdk) - [Claude Code](https://composio.dev/toolkits/gemini/framework/claude-code) - [Claude Cowork](https://composio.dev/toolkits/gemini/framework/claude-cowork) - [Codex](https://composio.dev/toolkits/gemini/framework/codex) - [Cursor](https://composio.dev/toolkits/gemini/framework/cursor) - [VS Code](https://composio.dev/toolkits/gemini/framework/vscode) - [OpenCode](https://composio.dev/toolkits/gemini/framework/opencode) - [OpenClaw](https://composio.dev/toolkits/gemini/framework/openclaw) - [Hermes](https://composio.dev/toolkits/gemini/framework/hermes-agent) - [CLI](https://composio.dev/toolkits/gemini/framework/cli) - [Google ADK](https://composio.dev/toolkits/gemini/framework/google-adk) - [LangChain](https://composio.dev/toolkits/gemini/framework/langchain) - [Mastra AI](https://composio.dev/toolkits/gemini/framework/mastra-ai) - [LlamaIndex](https://composio.dev/toolkits/gemini/framework/llama-index) - [CrewAI](https://composio.dev/toolkits/gemini/framework/crew-ai) ## Related Toolkits - [Composio](https://composio.dev/toolkits/composio) - Composio is an integration platform that connects AI agents with hundreds of business tools. It streamlines authentication and lets you trigger actions across services—no custom code needed. - [Composio search](https://composio.dev/toolkits/composio_search) - Composio search is a unified web search toolkit spanning travel, e-commerce, news, financial markets, images, and more. It lets you and your apps tap into up-to-date web data from a single, easy-to-integrate service. - [Perplexityai](https://composio.dev/toolkits/perplexityai) - Perplexityai delivers natural, conversational AI models for generating human-like text. Instantly get context-aware, high-quality responses for chat, search, or complex workflows. - [Browser tool](https://composio.dev/toolkits/browser_tool) - Browser tool is a virtual browser integration that lets AI agents interact with the web programmatically. It enables automated browsing, scraping, and action-taking from any AI workflow. - [Ai ml api](https://composio.dev/toolkits/ai_ml_api) - Ai ml api is a suite of AI/ML models for natural language and image tasks. It provides fast, scalable access to advanced AI capabilities for your apps and workflows. - [Aivoov](https://composio.dev/toolkits/aivoov) - Aivoov is an AI-powered text-to-speech platform offering 1,000+ voices in over 150 languages. Instantly turn written content into natural, human-like audio for any application. - [All images ai](https://composio.dev/toolkits/all_images_ai) - All-Images.ai is an AI-powered image generation and management platform. It helps you create, search, and organize images effortlessly with advanced AI capabilities. - [Anthropic administrator](https://composio.dev/toolkits/anthropic_administrator) - Anthropic administrator is an API for managing Anthropic organizational resources like members, workspaces, and API keys. It helps you automate admin tasks and streamline resource management across your Anthropic organization. - [Api labz](https://composio.dev/toolkits/api_labz) - Api labz is a platform offering a suite of AI-driven APIs and workflow tools. It helps developers automate tasks and build smarter, more efficient applications. - [Apipie ai](https://composio.dev/toolkits/apipie_ai) - Apipie ai is an AI model aggregator offering a single API for accessing top AI models from multiple providers. It helps developers build cost-efficient, latency-optimized AI solutions without juggling multiple integrations. - [Astica ai](https://composio.dev/toolkits/astica_ai) - Astica ai provides APIs for computer vision, NLP, and voice synthesis. Integrate advanced AI features into your app with a single API key. - [Bigml](https://composio.dev/toolkits/bigml) - BigML is a machine learning platform that lets you build, train, and deploy predictive models from your data. Its intuitive interface and robust API make machine learning accessible and efficient. - [Botbaba](https://composio.dev/toolkits/botbaba) - Botbaba is a platform for building, managing, and deploying conversational AI chatbots across messaging channels. It streamlines chatbot automation, making it easier to integrate AI into customer interactions. - [Botpress](https://composio.dev/toolkits/botpress) - Botpress is an open-source platform for building, deploying, and managing chatbots. It helps teams automate conversations and deliver rich, interactive messaging experiences. - [Chatbotkit](https://composio.dev/toolkits/chatbotkit) - Chatbotkit is a platform for building and managing AI-powered chatbots using robust APIs and SDKs. It lets you easily add conversational AI to your apps for better user engagement. - [Cody](https://composio.dev/toolkits/cody) - Cody is an AI assistant built for businesses, trained on your company's knowledge and data. It delivers instant answers and insights, tailored for your team. - [Context7 MCP](https://composio.dev/toolkits/context7_mcp) - Context7 MCP delivers live, version-specific code docs and examples right from the source. It helps developers and AI agents instantly retrieve authoritative programming info—no more out-of-date docs. - [Customgpt](https://composio.dev/toolkits/customgpt) - CustomGPT.ai lets you build and deploy chatbots tailored to your own data and business needs. Get precise and context-aware AI conversations without writing code. - [Datarobot](https://composio.dev/toolkits/datarobot) - Datarobot is a machine learning platform that automates model development, deployment, and monitoring. It empowers organizations to quickly gain predictive insights from large datasets. - [Deepgram](https://composio.dev/toolkits/deepgram) - Deepgram is an AI-powered speech recognition platform for accurate audio transcription and understanding. It enables fast, scalable speech-to-text with advanced audio intelligence features. ## Frequently Asked Questions ### What are the differences in Tool Router MCP and Gemini MCP? With a standalone Gemini MCP server, the agents and LLMs can only access a fixed set of Gemini tools tied to that server. However, with the Composio Tool Router, agents can dynamically load tools from Gemini and many other apps based on the task at hand, all through a single MCP endpoint. ### Can I use Tool Router MCP with Vercel AI SDK v6? Yes, you can. Vercel AI SDK v6 fully supports MCP integration. You get structured tool calling, message history handling, and model orchestration while Tool Router takes care of discovering and serving the right Gemini tools. ### Can I manage the permissions and scopes for Gemini while using Tool Router? Yes, absolutely. You can configure which Gemini scopes and actions are allowed when connecting your account to Composio. You can also bring your own OAuth credentials or API configuration so you keep full control over what the agent can do. ### How safe is my data with Composio Tool Router? All sensitive data such as tokens, keys, and configuration is fully encrypted at rest and in transit. Composio is SOC 2 Type 2 compliant and follows strict security practices so your Gemini data and credentials are handled as safely as possible. --- [See all toolkits](https://composio.dev/toolkits) · [Composio docs](https://docs.composio.dev/llms.txt)