# NVIDIA Vision MCP > NVIDIA Vision connects powerful visual APIs to your AI client, letting you generate images from text prompts or analyze existing visuals. Use it to ask questions about photos, detect objects in complex scenes, or extract data from scanned documents and forms. It handles everything from artistic style transfers to detailed business understanding. ## Overview - **Category:** industry-titans - **Price:** Free - **Tags:** computer-vision, image-generation, object-detection, visual-qa, image-captioning, generative-ai ## Description This MCP lets you treat images like structured data. Instead of manually running through different services—one for object counting, another for captioning, and a third for document reading—you just ask your agent a question about the image. You can generate brand-new concepts using Stable Diffusion models based only on text prompts, or feed it a scanned receipt and have it pull out the total amount due and the vendor name. When you subscribe through Vinkius, your AI client gets access to this entire suite of visual tools in one place. It’s built for professionals who need deep understanding from visuals, whether they are creating marketing assets or analyzing financial records. ## Tools ### image_captioning Generates a descriptive text summary detailing the contents and context of an image. ### detect_objects Identifies and provides a list of every physical object present in an uploaded picture. ### document_qa Reads scanned documents, forms, or receipts and answers specific questions about the contained text and data. ### generate_image Creates a brand-new image file from scratch based on a written text prompt using Stable Diffusion models. ### visual_grounding Pinpoints and isolates specific objects or phrases within an image, telling you exactly where they are located. ### image_segmentation Separates an image into distinct regions, allowing you to identify and isolate every major object present. ### style_transfer Applies the artistic look or style of one picture onto another existing visual asset. ### list_vision_models Retrieves a list of all available vision models that can be used with the NVIDIA API Catalog. ### visual_question_answering Allows you to ask natural language questions about an image and receive a direct answer based on its visual content. ## Prompt Examples **Prompt:** ``` Generate an image of a futuristic city at sunset. ``` **Response:** ``` Image generated successfully! Base64 data available for display. ``` **Prompt:** ``` What objects do you see in this image: https://example.com/photo.jpg ``` **Response:** ``` I detect: 1. A red car (center). 2. A tree (left). 3. A building (background). 4. Two people walking (right). ``` **Prompt:** ``` Describe this image in detail: https://example.com/document.png ``` **Response:** ``` The image shows a business document dated March 2026. It contains a table with revenue figures totaling $2.4M. ``` ## Capabilities ### Create new images from text Generate high-quality, unique images instantly using Stable Diffusion models based on detailed written descriptions. ### Answer questions about visuals Upload a photo and ask specific questions; the agent reads the image content and provides a detailed answer. ### Extract data from documents Process scanned forms, receipts, or business papers to accurately identify and pull out key pieces of information. ### Identify objects in images List every object visible in a picture, or locate specific items within the frame using visual grounding. ### Describe image contents Get rich, detailed captions that summarize everything happening in an image without needing to ask follow-up questions. ## Use Cases ### Analyzing competitor product shots A market researcher uploads multiple photos of competing products. They use detect_objects to count the number of visible features (like ports or buttons) and then use visual_question_answering to confirm if a specific brand logo is present on each device. ### Processing old legal contracts A paralegal receives dozens of poorly scanned, handwritten agreements. They feed the batch into document_qa, which accurately reads and extracts key clauses like 'Effective Date' and 'Termination Clause', saving hours of manual transcription. ### Designing a mood board for a client A designer is stuck on a concept. They use generate_image to create several visual options—like 'a brutalist building covered in moss' or 'futuristic beach at twilight'—and then uses image_segmentation to isolate key elements from the best result. ### Cataloging scientific research photos A biologist uploads images of local flora. They use detect_objects to list all visible species and run visual_grounding to pinpoint exactly where specific plant parts (like seeds or root systems) are located in the photo. ## Benefits - Stop guessing what an image means. Use visual_question_answering to ask your agent specific questions about any photo—like 'What brand is this watch?' or 'When did this meeting happen?' and get a definitive answer. - Never start from scratch again. The generate_image tool lets you build marketing concepts instantly, simply by typing out what you need in a text prompt, skipping the initial brainstorming phase entirely. - Process paperwork faster than ever. Instead of manually reading tables on scanned receipts, document_qa extracts figures like tax IDs and subtotals into clean data points you can use immediately. - Gain visual control over your assets. The style_transfer tool lets a designer take an existing photo and make it look like a Renaissance painting or a cyberpunk graphic with one command. - Improve searchability of visuals. Use image_captioning to get detailed, searchable descriptions for every photo you upload, making large archives instantly discoverable. ## How It Works The bottom line is that your AI client can seamlessly switch between creating visual content and deeply understanding existing images, all through one connection. 1. Subscribe to this MCP and provide your API Key from NVIDIA's developer site. 2. Direct your AI client (like Cursor or Claude) to the visual task, providing either a text prompt or an image URL. 3. Your agent uses the appropriate tool—whether it’s generating an asset or analyzing data—and returns the structured result directly to you. ## Frequently Asked Questions **Can I use NVIDIA Vision to generate images for a website?** Yes, absolutely. You use the generate_image tool by providing a text prompt (e.g., 'minimalist corporate office') and selecting your desired model parameters. **Does NVIDIA Vision help with legal documents?** It does. The document_qa tool is specifically designed to work with scanned forms, receipts, and contracts, allowing you to ask questions about the text it finds inside. **What is the difference between image_captioning and visual_question_answering?** Image captioning provides a general description of everything in an image. Visual question answering requires you to ask a specific query, like 'Who is this person?' or 'What year was this built?' for a targeted answer. **Do I need a developer background to use NVIDIA Vision?** No. You connect the MCP using your API key, but after that, you interact with it through natural conversation via your AI client, which handles all the complex coding for you. **Can I isolate specific parts of an image using NVIDIA Vision?** Yes. You can use visual_grounding to pinpoint a specific object or phrase and image_segmentation to cleanly separate that object from the rest of the picture.