Frequently Asked Questions about the MCP Server for Replicate

Q: What is the MCP Server? A: The MCP (Model Context Protocol) Server is an implementation for Replicate’s API that allows you to run Replicate models through a simple tool-based interface. It acts as a bridge between Replicate’s models and applications like Claude Desktop, Cursor, Cline, and Continue.

Q: What is Replicate? A: Replicate is a platform that allows you to run open-source machine learning models in the cloud. It provides a simple API for running models and allows you to easily integrate them into your applications.

Q: How do I get a Replicate API token? A: You can get a Replicate API token by visiting https://replicate.com/account/api-tokens and creating a new token.

Q: How do I configure Claude Desktop to use the MCP Server? A: Open Claude Desktop Settings (⌘,), select the “Developer” section, click “Edit Config”, and add the configuration block provided in the Quickstart section, replacing your_token_here with your Replicate API token.

Q: Can I use the MCP Server with other MCP clients besides Claude Desktop? A: Yes, you can use the MCP Server with any MCP client, such as Cursor, Cline, or Continue. You’ll need to configure the client with the server’s address and your Replicate API token.

Q: What models can I use with the MCP Server? A: You can use any model available on Replicate with the MCP Server.

Q: What kind of inputs can I use with the models? A: You can use text or structured input, depending on the requirements of the specific model you are using.

Q: How do I track the status of a prediction? A: You can use the get_prediction tool to check a prediction’s status. The MCP Server also provides real-time updates on the prediction’s progress.

Q: Can I cancel a running prediction? A: Yes, you can use the cancel_prediction tool to stop a running prediction.

Q: What are the available tools in the MCP Server? A: The MCP Server provides tools for model search, browsing, and information retrieval (search_models, list_models, get_model, list_collections, get_collection), prediction management (create_prediction, create_and_poll_prediction, get_prediction, cancel_prediction, list_predictions), and image handling (view_image, clear_image_cache, get_image_cache_stats).

Q: How do I troubleshoot issues with the MCP Server? A: Check that Claude Desktop is properly configured, ensure your Replicate API token is set correctly, restart the server and Claude Desktop, and check the server logs for any error messages.

Q: What are the requirements for using the MCP Server? A: You need Node.js >= 18.0.0, TypeScript >= 5.0.0, and Claude Desktop (or another MCP client) to use the tools.

Q: Where can I find more information about the Model Context Protocol (MCP)? A: You can find more information about MCP at https://github.com/mcp-sdk/mcp.

Q: How does the MCP Server integrate with UBOS? A: The MCP Server, when used within the UBOS platform, allows you to centralize AI agent management, seamlessly integrate data, develop custom AI agents, orchestrate multi-agent systems, and enhance security and governance.