comfyui-video

Give your AI agent a local video generation engine.

No API keys. No cloud bills. No queue times. Your agent generates videos directly on your Mac using ComfyUI — the same engine professionals use.

What It Does

• Text → Video in one tool call. Your agent calls generate_video(prompt="..."), waits for completion, and gets back a file URL. Done.
• Runs 100% locally. ComfyUI runs on your machine. Your prompts, your outputs, zero data sent to the cloud.
• Full agent control via MCP. 9 tools covering generation, job management, model listing, and system monitoring — everything an agent needs to operate ComfyUI autonomously.

Who This Is For

• Developers building AI pipelines that need video output
• Creators who want their AI agent to generate video content on demand
• Researchers running local AI experiments without cloud cost concerns
• Anyone using Claude/OpenClaw who wants a video generation capability that just works

Requirements

• Mac with Apple Silicon (M1–M4, tested on M2 Max)
• ComfyUI v0.16+ installed at ~/ComfyUI
• Wan 2.1 T2V models (~10GB total)
• Python 3.12+ for the MCP server
• OpenClaw or Claude Desktop with MCP support
• 16GB+ unified memory (32GB recommended)

The 9 MCP Tools

| Tool | What It Does | |------|-------------| | health_check() | Verify ComfyUI is reachable, GPU/VRAM status | | generate_video(...) | Queue a T2V job (returns job_id immediately) | | generate_image(...) | Queue a T2I job for still frames | | get_job_status(job_id) | Non-blocking status check | | wait_for_job(job_id) | Block until complete, returns output URLs | | list_models(type) | List available models by category | | queue_workflow(json) | Submit raw ComfyUI workflow JSON | | get_system_stats() | GPU utilization, VRAM usage, queue depth | | cancel_job(job_id) | Cancel a running or queued job |

$29 one-time — buy once, use forever. No subscription, no per-generation fees.

Built by ClawGear.