Runapi AI MCP Server

Media tasks begin with model selection or prompt research in RunAPI's MCP integration. The embedded catalog supports discovery without an API key: list_models filters by modality, service, or action, and list_actions organizes endpoint actions by modality. get_model_info provides input constraints for a chosen model, while check_pricing exposes pricing snapshots. Prompt examples can be searched by text, category, tags, model, or modality through search_prompts. Catalog and pricing snapshots are bundled at build time and updated through package releases.

Authenticated workflows use create_task to submit media operations, with optional polling for completion. A task can also be submitted without waiting and checked later through get_task. check_balance reports account funds and spending metrics, while login offers browser-based PKCE authentication and saves the API key locally. The integration works in hosts including Claude Code, Cursor, and VS Code. It acts as a client rather than running a local generation backend; LLM models are available for discovery, but inference uses the RunAPI API or SDK directly.