# Scale AI MCP for AI Agents AI Agent Connect > Scale AI MCP lets you manage data labeling and RLHF projects directly from your AI agent. Create projects, organize batches, and submit multi-modal annotation tasks like image segmentation or video playback. It handles the lifecycle of your training data, from initial setup to final batch completion. ## Overview - **Category:** ai-frontier - **Price:** Free - **Endpoint:** https://edge.vinkius.com/vk_preview_lsC40U9ook5OZ8hrmwvYLqntMTpGHdSfQWvhNSRl/ai-agent-connect - **Tags:** data-labeling, rlhf, machine-learning, annotation, training-data, project-management ## Description Scale AI gives you a direct line to your data labeling pipelines without switching tabs between your chat interface and a complex dashboard. Instead of manually clicking through project settings or dragging files into batches, you just tell your agent what needs to happen. You can spin up new projects for semantic segmentation, submit large batches of images for annotation, or tweak project parameters on the fly to fix quality issues. It handles the heavy lifting of orchestrating high-volume work for RLHF or custom model training. When you use this Connector, the friction of data preparation disappears. You can describe a project's goals in plain English, and your agent will handle the technical setup in the background. If the labeling team needs a change in instructions, you don't have to find the right button in a dashboard; you just ask your agent to update the project parameters. This keeps your workflow moving and ensures your training data stays consistent. This MCP is part of the Vinkius catalog, making it easy to plug into your existing workflow so you can focus on the actual science of machine learning rather than the logistics of data collection. It turns a multi-step administrative process into a simple conversation. ## Tools ### cancel_task Stop a pending task immediately. This helps you stay on budget by preventing unnecessary costs for data that's no longer needed. ### create_batch Group your data into a new batch. This allows you to organize high-volume work and trigger the labeling process for many items at once. ### create_image_annotation_task Submit a specific task for image annotation. This lets you send individual images to your labeling queue for specific projects. ### create_named_entity_recognition_task Create a task for identifying and extracting entities from text. Use this to pull specific names or locations from large datasets. ### create_project Set up a new Scale AI project with specific labeling configurations. This is the first step in starting a new data collection or RLHF run. ### create_segment_annotation_task Submit a task for semantic segmentation of images or other media. Use this for complex pixel-level labeling jobs. ### create_text_collection_task Create a task for gathering and labeling text data. This is useful for building large-scale text datasets for model training. ### create_video_playback_annotation_task Submit a task for video annotation and playback. This lets you send video files for temporal or object-based labeling. ### finalize_batch Complete a batch to trigger the actual labeling process. Use this to move your organized data into the production labeling queue. ### get_task Retrieve the status and details of a specific annotation task. This helps you monitor progress and keep your project on schedule. ### update_project_params Change the instructions or configuration for an existing project. This lets you refine labeling quality without restarting the entire project. ## Prompt Examples **Prompt:** ``` Create a new image annotation project called 'Street-Signs' for imageannotation. ``` **Response:** ``` I've created the project 'Street-Signs'. You can now start creating batches or submitting tasks to this project. ``` **Prompt:** ``` Submit an image annotation task to project 'Street-Signs' with the image URL 'https://example.com/sign.jpg'. ``` **Response:** ``` Task created successfully. The Task ID is 'task_abc123'. You can track its progress using the get_task tool. ``` **Prompt:** ``` Finalize the batch named 'sprint-01-batch'. ``` **Response:** ``` Batch 'sprint-01-batch' has been finalized. Scale will now begin processing the tasks within this batch. ``` ## Capabilities ### Create labeling projects Start new projects for tasks like image annotation or semantic segmentation. ### Manage high-volume batches Group your data into manageable work units to trigger the labeling process. ### Submit multi-modal tasks Send images, videos, and text collections for various annotation types. ### Track task status Check the progress of individual items or cancel pending ones to stay on budget. ### Tune project parameters Update instructions and settings dynamically to improve labeling quality. ## Use Cases ### Image Segmentation Project An ML engineer needs to label 5,000 images. They ask the agent to create a project and submit the images as a batch using create_batch. ### RLHF Tuning A researcher wants to update instructions for a human feedback loop. They tell the agent to change the project parameters immediately with update_project_params. ### Video Annotation A developer needs to tag video clips. The agent uses the create_video_playback_annotation_task tool to submit the files to the queue. ### Budget Management A manager sees a stuck batch and asks the agent to cancel specific tasks to save on costs using cancel_task. ## Benefits - Faster project setup: Use create_project to launch new labeling jobs in seconds rather than navigating a web UI. - Simplified batching: Group your data quickly with create_batch to keep your labeling pipeline organized. - Dynamic quality control: Use update_project_params to fix labeling errors on the fly without restarting projects. - Real-time tracking: Check the status of every item using get_task to monitor your budget and throughput. - Multi-modal support: Handle images, videos, and text with tools like create_video_playback_annotation_task. ## How It Works The bottom line is you can manage your entire Scale AI data pipeline through a simple conversation. 1. Connect your Scale AI Live API Key to your agent. 2. Describe the labeling project or batch you need to create. 3. Get confirmation and start managing your training data pipeline. ## Frequently Asked Questions **What can I do with the Scale AI MCP?** You can manage your entire data labeling pipeline through a conversation. This includes creating projects, organizing batches, and submitting various annotation tasks like image segmentation or video playback. **How does the Scale AI MCP help with RLHF?** It allows you to quickly spin up RLHF projects and update project parameters on the fly. This makes it much easier to refine instructions for human feedback loops without manual dashboard updates. **Can I use the Scale AI MCP for video labeling?** Yes, the Connector includes specific tools for video playback annotation. You can submit video files to the queue and have them processed for temporal or object-based labeling. **Does the Scale AI MCP support semantic segmentation?** Yes, it supports semantic segmentation tasks. You can submit these tasks directly through your AI agent to get pixel-level labeling for your training data. **How do I manage my budget with the Scale AI MCP?** You can monitor your spending by checking the status of individual tasks. If you need to stop work on a specific item, you can cancel pending tasks to avoid unnecessary costs. **Can I update project instructions with the Scale AI MCP?** Yes, you can update project parameters dynamically. If your labeling requirements change, just tell your agent to update the instructions for the active project. **How do I start a high-volume labeling job using batches?** First, use `create_batch` to initialize a group for your project. After submitting your tasks to this batch, call `finalize_batch` to signal Scale to begin the labeling process. **Can I check the status of a specific annotation task?** Yes, use the `get_task` tool with the specific Task ID. It will return the full metadata, current status, and any available results for that unit of work. **What should I do if I submitted a task by mistake?** You can use the `cancel_task` tool with the Task ID. If you need to reuse the unique identifier, you can also set the `clear_unique_id` parameter to true.