short-video-agent-kit

Purpose

Short Video Agent Kit is a short video generation toolkit for AI agents, providing a unified CLI command-line tool and MCP server, allowing agents to build, validate, and (when explicitly enabled) generate vertical short videos through a single interface. It supports Sora/OpenAI, Gemini Veo, xAI/Grok, and Seedance/PiAPI-style providers, and is part of the Delx agent technology stack.

Main Capabilities

• Unified interface across multiple providers: Provides a consistent calling method for MCP clients such as Codex, Claude, Cursor, Hermes, and OpenClaw, covering video generation providers such as Sora, Veo, xAI, and Seedance.
• Dry-run safe mode by default: Can be tried without an API key; the tool returns the complete provider-agnostic generation plan it would send, without consuming any quota.
• Payload construction and validation: Agents can first build prompt-to-video request payloads, inspect parameter shapes, and re-target endpoints and parameter structures across different providers.
• Provider readiness check: Can detect which providers have configured keys (keys are only detected and are not printed or returned).
• Privacy boundary: The tool returns privacy boundary information; prompts and outputs are stored in user-controlled local paths.
• Paid generation requires explicit enablement: Only after using --live, MCP's live=true , or setting SHORT_VIDEO_DRY_RUN=false will it actually call the provider for paid generation.
• Transport methods: Uses stdio by default, with optional Streamable HTTP (no session ID, JSON responses, loopback address only), configured via the environment variable SHORT_VIDEO_MCP_HOST、SHORT_VIDEO_MCP_PORT、SHORT_VIDEO_MCP_TRANSPORT=http .

Usage

After installation, you can run the CLI directly or connect it as an MCP server to an agent client. You can try dry-run without an API key. For example, when building a Sora plan for an 8-second vertical trailer, the tool returns real output (without calling the provider or consuming quota). When switching the same prompt to a different provider, the plan automatically re-targets the endpoint and parameter shape. When checking provider configuration status, keys are only detected and not printed. When ready for actual rendering, set the provider key and rerun generate using --live(or SHORT_VIDEO_DRY_RUN=false).

Recommended first calls include:short_video_connection_status、short_video_privacy_audit、short_video_build_payload、short_video_generate.

For configuration, you can copy .env.example as .env, and fill in only the provider keys you plan to use;.env、output/ and .agent-data/ are both ignored by Git.

Use Cases

Suitable for scenarios where agents need prompt-to-video payload validation across multiple video providers, as well as optional paid generation. Especially suitable for agent workflows that want to safely check provider readiness, build generation plans, and confirm privacy boundaries before actually consuming quota.

Limitations

• Paid generation must be explicitly enabled; by default, providers are not called.
• Reference images must be user-owned or authorized.
• Outputs are written to user-controlled local paths.
• API keys are not returned by the tool.
• HTTP transport is limited to loopback addresses and has no session ID.

Contact and Support

• Support email: support@delx.ai (general questions, integration help, partnerships)
• Bug reports / feature requests: GitHub Issues
• Updates: @delx369 on X
• Website: wellness.delx.ai