Replicate MCP

Replicate MCP is an implementation of the Model Context Protocol that lets AI agents and tools access Replicate's catalog of open-source models through a standardized interface, without manual API integration.

Claude Desktop, Claude Code, Cursor, GitHub Copilot, and any client that implements the Model Context Protocol can connect to Replicate's MCP server and run models.

Add the Replicate MCP server URL and your API token to your MCP client configuration. The client will automatically discover available models and expose them as callable tools.

Replicate hosts hundreds of models including Flux, Stable Diffusion, LLaMA, Whisper, and specialized models for image generation, video, audio, and text processing.

The MCP server itself is free and open source. You pay Replicate's standard per-run pricing for each model execution, which varies by model and compute requirements.

Yes. In Wireflow you can connect multiple models on the visual canvas, passing outputs from one model as inputs to the next, creating multi-step generation pipelines.

MCP adds automatic tool discovery, so agents find and use models without pre-coded integrations. The REST API requires explicit endpoint calls and response parsing for each model.

Yes. Build a workflow on the Wireflow canvas using Replicate models, then expose it as an API endpoint. External apps can trigger the full pipeline with a single HTTP call.