From a creative brief to a short video: a Medux MCP workflow

I'm Fenii, part of the Medux team. If you're building a video agent, here's a small workflow to start with: generate a few assets, review the results, and assemble a short clip.

Watch our product demo:

1. Connect your agent to Medux MCP. The setup guide includes Codex with OAuth and other supported clients: https://medux.io/overview/mcp

2. Start with a specific brief and approved reference images. For example: create two short product scenes and a short voiceover. Generate or reuse the visual assets, then animate an image with Image to Video: https://medux.io/playground/video/image-to-video

3. Check each task's status and result before moving on. Review the scenes and voice, then use the returned assets in the next step rather than starting from scratch.

4. Assemble the approved clips with Merge Video: https://medux.io/playground/video/merge-video . If you also need music or voice, check the inputs and outputs of the relevant audio step separately.

A starter brief to adapt: "Use these two approved images to make a short product clip. Show me the generated scenes before assembling the final video."

Which step would you want your agent to handle first: scene generation, voice, or final assembly?

Quick start for your first test

Prepare one reference image you own or have permission to use, plus a short brief describing the movement and intended result. Start with one scene and a short duration available for your selected model. Music or voice is optional for this first test; use authorized audio if you add it later.

Browser route: sign in and open Image to Video: https://medux.io/playground/video/image-to-video . Choose the model and settings, review the credit estimate, then submit one task. Check the result before adding more steps.

Agent route: follow the Codex OAuth setup at https://medux.io/overview/mcp/codex . In the capability page, switch to For AI Agents and check the MCP tool mapping. Ask the agent to create one scene, return the task ID, check its status, and show the result for your review before continuing.

If something fails, reply here with the tool/model, requested duration and resolution, client/browser, approximate time with timezone, task ID if available, and the exact error. Tell us which step failed and what you expected. A redacted screenshot helps. Please don't post API keys, tokens, account details, or private source media.

If you need to share sensitive diagnostic details, say so here and we can arrange a private follow-up. Please check the original task status before resubmitting, so a retry doesn't accidentally create another job. We will check the task and billing details rather than assume every failure has the same cause.