Short Video Agent Kit
This MCP server provides an agent-first toolkit for generating short-form AI videos across multiple providers (OpenAI Sora, Gemini Veo, xAI Grok, Seedance/PiAPI) with a dry-run-by-default safety approach. You can:
Get an agent manifest (
short_video_agent_manifest) with machine-readable install, provider, runtime, and safety guidance tailored to clients like Codex, Claude, Cursor, Hermes, and OpenClaw.Check provider connection status (
short_video_connection_status) to see which providers are configured and ready, without exposing API keys, and confirm dry-run mode.Run a privacy audit (
short_video_privacy_audit) covering prompts, reference assets, output paths, and API keys.Build provider-specific payloads (
short_video_build_payload) with prompt, duration, aspect ratio, resolution, size, model, negative prompt, reference images, and output path—without making paid calls.Generate short-form vertical videos (
short_video_generate) with the same parameters, using live generation only when explicitly enabled (live=trueorSHORT_VIDEO_DRY_RUN=false); otherwise returns a safe dry-run plan.Operate as a CLI tool or streamable HTTP server, with JSON or Markdown response formats.
⭐ If this agent-first tool helps your workflow, please star the repo. Stars make this tooling easier for other builders to discover and help Delx keep shipping open infrastructure. 🧱 Part of the Delx agent stack — 15 open-source MCP servers across body, reach and coordination.
HTTP (v2 stateless)
Default is stdio. Optional Streamable HTTP — no session id, JSON responses, loopback only:
npx -y -p short-video-agent-kit short-video-mcp --http
# GET http://127.0.0.1:3033/health
# POST http://127.0.0.1:3033/mcp (sessionless)Env: SHORT_VIDEO_MCP_HOST, SHORT_VIDEO_MCP_PORT, SHORT_VIDEO_MCP_TRANSPORT=http.
Provider-neutral short-form AI video toolkit for agents. It gives Codex, Claude, Cursor, Hermes, OpenClaw and other MCP clients one interface for building dry-run payloads and, when explicitly enabled, generating vertical video through Sora/OpenAI, Gemini Veo, xAI/Grok and Seedance/PiAPI-style providers.
Use it when an agent needs one safe interface for prompt-to-video payload validation and optional paid generation across multiple providers.
Related MCP server: ai-promo-video
Why It Is Agent-First
Video generation can be expensive and prompt-sensitive. This package makes agents start with safe steps:
inspect provider readiness
return privacy boundaries
build payloads without spending credits
require
--liveorSHORT_VIDEO_DRY_RUN=falsebefore provider callskeep prompts and outputs in local user-controlled paths
Install
npm install -g short-video-agent-kitOr run directly:
npm exec --yes --package=short-video-agent-kit -- short-video-agent-kit doctorQuickstart
No API key needed to try it — generation is dry-run by default, so the kit returns the exact provider-neutral plan it would send without spending a credit.
Build a Sora plan for an 8-second vertical teaser:
short-video-agent-kit generate \
--provider openai_sora \
--prompt "Vertical 8-second product teaser for a minimalist water bottle, soft studio light, slow dolly-in" \
--output ./output/teaser.mp4Real output (no provider call, no credits spent):
{
"ok": true,
"dry_run": true,
"next_step": "Pass --live or set SHORT_VIDEO_DRY_RUN=false to call the provider API.",
"provider": "openai_sora",
"endpoint": "POST /v1/videos",
"payload": {
"model": "sora-2",
"prompt": "Vertical 8-second product teaser for a minimalist water bottle, soft studio light, slow dolly-in",
"seconds": "8",
"size": "720x1280"
}
}Same prompt, different provider — the plan re-targets the endpoint and parameter shape for you. payload returns just the plan (no dry_run wrapper):
short-video-agent-kit payload --provider gemini_veo --prompt "Same teaser, 9:16, cinematic"{
"provider": "gemini_veo",
"endpoint": "POST /models/{model}:predictLongRunning",
"payload": {
"instances": [
{
"prompt": "Same teaser, 9:16, cinematic"
}
],
"parameters": {
"aspectRatio": "9:16",
"durationSeconds": 8
}
}
}Check which providers are wired up (keys are detected, never printed):
short-video-agent-kit doctor{
"ok": false,
"dry_run": true,
"providers": {
"openai_sora": { "configured": false, "env_keys": ["OPENAI_API_KEY"], "models": ["sora-2"] },
"gemini_veo": { "configured": false, "env_keys": ["GEMINI_API_KEY", "GOOGLE_API_KEY"], "models": ["veo-3.1-fast-generate-preview"] },
"xai_grok": { "configured": false, "env_keys": ["XAI_API_KEY"], "models": ["grok-imagine-video", "grok-imagine-image"] },
"seedance_piapi": { "configured": false, "env_keys": ["PIAPI_KEY", "SEEDANCE_API_KEY"], "models": ["seedance-2-fast-preview"] }
},
"output_dir": "./output",
"next_steps": [
"Set one provider key: OPENAI_API_KEY, GEMINI_API_KEY, XAI_API_KEY or PIAPI_KEY."
]
}When you are ready to actually render, set a provider key and re-run generate with --live (or SHORT_VIDEO_DRY_RUN=false).
CLI
short-video-agent-kit manifest --client codex
short-video-agent-kit doctor
short-video-agent-kit privacy-audit
short-video-agent-kit payload --provider gemini_veo --prompt-file prompt.txt
short-video-agent-kit generate --provider openai_sora --prompt "Vertical product teaser" --output ./output/teaser.mp4
short-video-agent-kit generate --provider openai_sora --prompt-file prompt.txt --output ./output/teaser.mp4 --liveSupported providers:
openai_soragemini_veoxai_grokseedance_piapi
MCP
short-video-mcpHTTP transport:
SHORT_VIDEO_MCP_TRANSPORT=http short-video-mcpHermes-style config:
mcp_servers:
short_video:
command: npx
args: ["-y", "short-video-agent-kit"]
sampling:
enabled: falseRecommended first calls:
short_video_connection_statusshort_video_privacy_auditshort_video_build_payloadshort_video_generate
Agent Surfaces
Tool | Purpose |
| Install/runtime guidance for Codex, Claude, Cursor, Hermes and OpenClaw |
| Provider readiness without API keys |
| Prompt, output and reference-asset boundaries |
| Provider-specific payload without paid generation |
| Dry-run by default, live only when explicitly requested |
Copy-Paste Agent Prompt
Use short-video-agent-kit. First call short_video_connection_status and short_video_privacy_audit.
Build the payload before generation. Only set live=true if I explicitly confirm a paid provider call.Configuration
Copy .env.example to .env and fill only the provider keys you plan to use. .env, output/ and .agent-data/ are ignored by Git.
Safety Model
Dry-run is the default.
API keys are never returned by tools.
Paid generation requires
--live, MCPlive=true, orSHORT_VIDEO_DRY_RUN=false.Reference images must be user-owned or licensed.
Outputs are written to local paths controlled by the user.
Development
npm install
npm test
npm run check📧 Contact & Support
📨 support@delx.ai — general questions, integration help, partnerships
🐛 Bug reports / feature requests — GitHub Issues
🐦 Updates — @delx369 on X
🌐 Site — wellness.delx.ai
Available Tools
5 toolsshort_video_agent_manifestShort Video Agent ManifestCRead-onlyIdempotent
Machine-readable install, provider, runtime and safety guidance for agents.
Inputs: client, response_format.
short_video_build_payloadBuild Short Video PayloadBRead-onlyIdempotent
Build a provider-specific request payload without calling the paid provider API.
Inputs: prompt (required), live, size, model, +10 more.
short_video_connection_statusShort Video Connection StatusBRead-onlyIdempotent
Check configured providers and dry-run mode without exposing API keys.
Inputs: response_format.
short_video_generateGenerate Short VideoB
Generate one vertical video. Returns a dry-run payload unless live=true or SHORT_VIDEO_DRY_RUN=false.
Inputs: prompt (required), live, size, model, +10 more.
short_video_privacy_auditShort Video Privacy AuditCRead-onlyIdempotent
Return prompt, asset, output and provider-key safety boundaries.
Inputs: response_format.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 5 tools
Each tool targets a distinct aspect of short video generation: configuration, payload construction, connectivity, generation, and privacy. No overlapping purposes.
All tools share the 'short_video_' prefix and follow a noun/verb_noun pattern, though 'generate' is a lone verb while others have two words. Minor deviation but overall clear.
5 tools is well-scoped for a focused agent kit covering setup, execution, and safety. No unnecessary tools.
Covers core lifecycle: metadata, payload, connectivity, generation, and privacy. Missing post-generation actions (e.g., list, delete) but acceptable for an agent kit.
Maintenance
Related MCP Connectors
AI video, images, music & SFX: Seedance 2.5, Veo 3.1, Kling 3.0, Nano Banana Pro, 20+ models.
AI video editor your agent can drive: plan scenes, edit a multi-track timeline, and render MP4s.
Video production studio for AI agents: AI media, motion graphics as code, timeline and export.
Agentic AI video editor: edit a real timeline, generate clips, music and voice, then render.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to edit videos through natural language, providing tools for timeline editing, audio management, rendering, and more.2MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to produce professional promo videos by providing tools for capture, media search, scene writing, rendering, editing, and incremental revision.12 npm17MIT
- AlicenseNot gradedqualityAmaintenanceLets any LLM agent actually watch videos: a watch_video tool takes a URL or local file and returns scene-aware keyframes fused with a timestamped transcript, processed 100% locally with per-source caching.2,328MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to autonomously edit videos into publish-ready vertical short-form content, including silence removal, subtitle generation, voiceover synthesis, color grading, and composite pipeline creation.MIT