Short Video Agent Kit

What davidmosiah/short-video-agent-kit MCP server does

davidmosiah/short-video-agent-kit MCP server provides an agent-oriented interface for short-form AI video workflows. It can inspect provider configuration, check privacy boundaries, translate a prompt into a provider-specific request payload, and optionally submit a live generation request. Supported provider identifiers are openai_sora, gemini_veo, xai_grok, and seedance_piapi.

The package is available through both a command-line interface and an MCP server. Its output is intended for vertical video use cases, with examples covering 8-second, 9:16 or 720x1280 content. Generated files are written to local paths selected by the user.

How it works

The normal workflow begins with readiness and safety checks. The doctor command, or the MCP connection-status tool, reports which provider credentials are configured without exposing their values. A privacy audit describes the boundaries around prompts, output locations, and reference assets.

Payload construction is separate from generation. The CLI payload command and the short_video_build_payload MCP tool return the request shape for a selected provider without making a paid API call. For example, Sora plans use a video endpoint with model, prompt, duration, and size fields, while Veo plans use a long-running prediction endpoint with instances and generation parameters.

Dry-run mode is the default. A live provider request requires the CLI --live option, MCP live=true, or SHORT_VIDEO_DRY_RUN=false. This makes davidmosiah/short-video-agent-kit MCP server suitable for agents that must prepare or validate requests before a user approves spending credits.

Setup and configuration

Install the package globally with npm, or invoke its commands through npm without a global installation. The MCP stdio entry point is short-video-mcp; Streamable HTTP can be enabled with SHORT_VIDEO_MCP_TRANSPORT=http, and the documented HTTP example listens on loopback at port 3033. Host and port can be adjusted with SHORT_VIDEO_MCP_HOST and SHORT_VIDEO_MCP_PORT.

Provider credentials are optional for dry-run use. Live operation can use OPENAI_API_KEY, GEMINI_API_KEY or GOOGLE_API_KEY, XAI_API_KEY, and PIAPI_KEY or SEEDANCE_API_KEY, depending on the selected provider. The README recommends copying .env.example to .env and filling only the keys required for the providers being used.

Tools and capabilities

The MCP surface includes:

• short_video_agent_manifest for runtime and client guidance.
• short_video_connection_status for provider readiness checks.
• short_video_privacy_audit for prompt, output, and reference-asset boundaries.
• short_video_build_payload for creating provider-specific plans without generation.
• short_video_generate for dry-run or explicitly approved live generation.

The CLI offers corresponding commands such as manifest, doctor, privacy-audit, payload, and generate. Prompts can be supplied directly or loaded from a file. The package also supports an HTTP MCP mode that returns stateless JSON responses on the documented local endpoints.

Limitations and notes

Live rendering depends on the selected provider's API access and may incur provider charges. The kit itself does not remove those provider requirements. API keys are detected but not returned by its tools, and dry-run plans do not produce video files.

Reference images must be owned by the user or properly licensed. Outputs remain in local, user-controlled paths. davidmosiah/short-video-agent-kit MCP server documents Codex, Claude, Cursor, Hermes, and OpenClaw as client environments, but the supplied material does not describe provider-specific feature parity beyond the payload examples.