# Replicate MCP for AI Agents AI Agent Connect > Replicate MCP lets your AI agent run, manage, and discover thousands of open-source ML models. Generate images, text, audio, and video directly through your chat interface. Track prediction statuses, browse curated collections, and check hardware specs without leaving your workspace. ## Overview - **Category:** ai-frontier - **Price:** Free - **Endpoint:** https://edge.vinkius.com/vk_preview_M4yBN2KNxSS8lU5z0YgsZ7knjRqsTTpfrLzeO7vW/ai-agent-connect - **Tags:** machine-learning, model-inference, generative-ai, api-integration, cloud-computing ## Description Replicate connects your account to any AI agent so you can run thousands of open-source ML models through natural conversation. This Connector turns your agent into a dedicated ML operations assistant. Instead of jumping between tabs to find a specific image generation model or checking if a video render finished, you just tell your agent what you need. It handles the heavy lifting of searching the catalog, finding the right model versions, and monitoring the progress of your jobs. You can see hardware pricing, explore curated categories like LLMs or text-to-video, and manage your usage in one place. By using this Connector through the Vinkius catalog, you get a direct line to the Replicate ecosystem. It removes the friction of manual model selection and status tracking, letting you focus on the output rather than the infrastructure. ## Tools ### get_account Verify your API token and check your current usage limits. It confirms your connection is active and healthy. ### get_collection Get specific details and featured models for a model category. This gives you a deeper look at a specific niche. ### get_model Get the full details and descriptions for a specific model slug. This is useful for checking the model's capabilities. ### get_model_versions List all version IDs for a model to ensure consistent results. Use this to lock in a specific version for production. ### get_prediction Check the status and output URLs of a specific prediction ID. Use this to see if your image or video is finished. ### list_collections Browse model groups like audio-to-audio or image-to-video. It shows you curated sets of related models. ### list_hardware See available GPU options and their pricing for inference. This helps you estimate costs before you start a run. ### list_models See a list of all available ML models and their requirements. This helps you find the right model for your specific task. ### list_predictions View your recent history of model runs and their outputs. Use this to keep track of all your past work. ### search_models Find specific models by typing a query like "text-to-image". It filters the catalog based on your keywords. ### cancel_prediction Stop a running model prediction immediately to save on costs. Use this if a job is taking too long or produced an error. ### create_prediction Run a specific model and get a prediction ID with its status. It handles the input schema so you don't have to guess the parameters. ## Prompt Examples **Prompt:** ``` Find some models that can turn text into video. ``` **Response:** ``` I found several collections and models for text-to-video. Here are the most popular options: | Model Name | Description | Hardware | | :--- | :--- | :--- | | **Model A** | High-quality cinematic video | A100 GPU | | **Model B** | Fast, stylized animation | L4 GPU | Would you like me to search for more specific details on one of these? ``` **Prompt:** ``` Run a model to make a photo of a cat in a space suit. ``` **Response:** ``` I've started that for you. **Prediction ID:** `pred_987654321` **Status:** `processing` I'll let you know as soon as the image is ready. Usually, this takes about 15 to 30 seconds. ``` **Prompt:** ``` Check my recent runs. ``` **Response:** ``` Here is the history of your recent model predictions: * **ID:** `pred_123` | **Status:** `succeeded` | **Output:** [Link to Image] * **ID:** `pred_456` | **Status:** `failed` | **Error:** Out of memory * **ID:** `pred_789` | **Status:** `processing` | **Model:** Llama-3-70B ``` ## Capabilities ### Run ML models Trigger image, text, or video generation directly from your chat. ### Track prediction status See if a model is still processing or if the result is ready. ### Browse model catalogs Search through thousands of models by name or category. ### Check hardware specs View GPU types and pricing for different inference workloads. ### Manage model versions Find specific version IDs for consistent model outputs. ### Monitor usage history View your recent predictions and account details. ## Use Cases ### Generating marketing images A designer asks the agent to create 10 variations of a product image. The agent uses `create_prediction` and notifies the user when they are ready. ### Testing LLM responses A developer wants to compare how different models handle a specific complex prompt. The agent searches the catalog and runs multiple tests. ### Batch video processing An editor needs to run a series of text-to-video tasks in a row. The agent tracks the history of all runs using `list_predictions`. ### Researching hardware An engineer checks GPU specs to see if a specific model fits their budget. They use `list_hardware` to compare pricing and specs. ## Benefits - Skip manual searches by using `search_models` to find the right tool instantly. - Save money by using `cancel_prediction` to stop jobs that aren't working. - Ensure consistency by selecting specific versions with `get_model_versions` for your production apps. - Monitor progress in real-time using `get_prediction` instead of refreshing your browser tabs. - Plan your budget by checking GPU pricing with `list_hardware` before you start a run. - Organize your projects by browsing curated groups with `list_collections` to see what's new. ## How It Works The bottom line is your AI agent handles the entire Replicate lifecycle from discovery to execution. 1. Connect your Replicate API token to the Connector. 2. Tell your AI agent which model to run and provide the inputs. 3. Get the results or URLs directly in your chat. ## Frequently Asked Questions **What models can I run with the Replicate MCP?** You can run thousands of open-source models, including those for image generation, text, audio, and video. Your agent can search the entire Replicate catalog to find exactly what you need. **How do I stop a model from running once it starts?** You can simply ask your agent to cancel the prediction. It will use the correct command to stop the job immediately, which helps you manage your costs. **Can I see how much my Replicate account has spent?** Yes, your agent can check your account details to show your current usage limits and information. This helps you keep track of your spending in real-time. **How do I find the best GPU for my specific model?** Your agent can list the available hardware options and their pricing. It will show you the specs for different GPUs so you can choose the best fit for your workload. **Can I run specific versions of a model for consistency?** Absolutely. Your agent can look up all available version IDs for a model. This allows you to lock in a specific version so your results remain consistent every time. **Is there a way to see all the different categories of models available?** Yes, your agent can list all the curated collections on Replicate. This lets you browse categories like 'text-to-image' or 'audio-to-audio' to find new tools. **How do I get a Replicate API token?** Log in to the [**Replicate API Tokens page**](https://replicate.com/account/api-tokens) and click **Create API Token**. Copy the token immediately — it starts with `r8_` and won't be shown again. **How do I run a model prediction?** Use `create_prediction` with the model slug (e.g. "stability-ai/sdxl") and an input JSON object matching the model's schema. The prediction starts as 'starting', then 'processing', and finally 'succeeded' with output URLs. Use `get_prediction` to check status and retrieve results. **How do I find models for specific tasks?** Use `search_models` with a query like 'text-to-image', 'llm', 'music-generation' or 'video-generation'. You can also use `list_collections` to browse curated collections by category, and `get_collection` to see featured models in each collection. **Can I cancel a running prediction?** Yes! Use `cancel_prediction` with the prediction ID. This works for predictions that are 'starting' or 'processing'. The status will change to 'canceled' and you won't be charged for the full compute time.