Replicate
Officialby Replicate
Replicate is an agent skill that lets AI coding agents discover, compare, and run thousands of open-source AI models through Replicate's cloud API. With a single install command, agents gain the ability to search curated model collections, fetch model schemas, create predictions, and retrieve generated outputs across image generation, text-to-speech, video creation, music composition, language models, and more. The skill teaches agents a structured workflow: choose the right model using search and collections APIs, validate inputs against each model's OpenAPI schema, create predictions via the HTTP API, poll for results or use webhooks, and return output URLs. It connects to Replicate's official MCP server for seamless tool integration and supports concurrent prediction execution for maximum throughput. Agents can leverage official models maintained by Replicate staff for stable interfaces and predictable pricing, or tap into community-contributed models for specialized tasks. The skill follows the open Agent Skills standard, making it compatible with Claude Code, OpenAI Codex, and other agent platforms.
Installation
Key Features
- ✓Search and discover AI models across curated collections via API
- ✓Run predictions on thousands of open-source models including image, audio, video, and text generation
- ✓Fetch and validate model input/output schemas automatically before creating predictions
- ✓Connect to Replicate MCP server for native tool integration with coding agents
- ✓Support concurrent predictions and webhook-based result delivery for efficient workflows
- ✓Access official models with stable APIs, always-on availability, and predictable pricing
Use Cases
- →Generate images, videos, and audio using state-of-the-art open-source AI models from within your agent workflow
- →Compare and evaluate multiple AI models for a specific task by querying Replicate's curated collections
- →Build AI-powered applications that run inference on demand with pay-per-prediction pricing
- →Automate media processing pipelines by chaining model predictions for tasks like image generation and upscaling
- →Prototype with community models and deploy production workloads on official always-on models