tolyapash/motionvid

O que motionvid faz?

Motionvid builds finished videos from a brief: script, animated scenes, voiceover, music and sound effects, rendered to MP4, from 5 seconds up to 20 minutes. You drive it through its MCP server. Act like a motion designer taking a brief, not a form: get the few facts that matter, make sensible choices for the rest, show the result, then iterate.

STOP: if Motionvid MCP tools (create_video, get_video_status, …) are available, use them. Do not hand-roll REST calls and do not write rendering code (Remotion, FFmpeg) yourself. Never ask the user to paste an API key into the chat.

1. Detect what's connected

1. Call whoami. If it answers, you are connected: go to step 2.
2. If the tools are missing, add the server and let the user sign in:
   • MCP endpoint: https://motionvid.ai/mcp (Streamable HTTP, OAuth 2.1 with discovery)
   • Claude Code: claude mcp add --transport http motionvid https://motionvid.ai/mcp, then claude mcp login motionvid
   • Codex: codex mcp add motionvid --url https://motionvid.ai/mcp, then codex mcp login motionvid
   • Cursor, VS Code, Claude, ChatGPT and others: https://motionvid.ai/agents Ask once ("Should I add Motionvid to your setup?") and do it on a yes.
3. Headless (CI, servers): the user creates an mvk_ key at https://motionvid.ai/account/api-keys and exposes it as an environment variable; send it as Authorization: Bearer <key>.
4. A "forbidden" or upgrade answer means the account has no paid plan: tell the user the connector is included on every paid plan (https://motionvid.ai/pricing) and stop.

2. Brief

Get, or decide yourself when the user doesn't care:

• What it's about and who it's for. Pass URLs, PDFs, images and audio as inputs: the planner reads pages, brand colours and documents itself.
• Length (duration, seconds) and shape (aspect_ratio: 16:9, 9:16, 1:1, 4:5).
• Sound: narration (and narration_language, narration_voice), music, sfx. If the user wrote the narration, put it in the prompt word for word and say so.
• Look: animation_style (list_animation_styles or suggest_animation_styles to let the user pick) and brand_kit_id (list_brand_kits).
• Mode: motion graphics by default. For a story, brand film or ad with people or a recurring character, use AI Director: ai_video: "full" (script, a consistent cast, cinematic scenes, voice and music). Add storyboard: true when the user wants to approve the cast and storyboard first. ai_video: "auto" mixes footage in only where the story needs it.

Optional: enhance_prompt turns a rough idea into a full brief.

3. Show the cost before spending

Call get_credit_balance (and list_models for per-model costs). A 30 second motion graphics video on the fast model typically takes about 33 credits, one minute about 45. AI Director films cost more because every scene is footage. For anything long (over 2 minutes) or AI Director, tell the user the rough cost from list_models and wait for a yes.

4. Create and follow

1. create_video with the brief. It returns a job id at once; the build runs in the background.
2. Poll get_video_status every 15 to 30 seconds. States: queued → processing → rendering → completed, or failed / canceled. A build usually takes a few minutes; slow is normal. Never cancel or restart because it looks idle.
3. awaiting_approval only happens when you asked for storyboard: true: call preview_video, show the user the board, and call approve_video with their decision (or changes). Never approve on their behalf.

5. Deliver and iterate

1. preview_video shows one frame per scene: show them to the user.
2. Changes: edit_video with the change in plain words ("make scene 3 punchier", "use our logo at the end"). Only that part is redone.
3. render_video for the MP4 (720p default; 1080p and 4K on plans that include them). Then get_video_status gives download_url: it expires after about 15 minutes, so download the file where the user asked (e.g. ./media/) or fetch a fresh link rather than storing one.
4. The user can also open and edit the video in the Motionvid app (app_url in the status).

Rules

• One video per request. Don't start a second job while the first is still running unless asked.
• Report readiness honestly: if a job failed, say so and quote its error; don't claim success.
• Keep the user's exact words for scripts, names, prices and claims.
• cancel_video only when the user asks.

When things break

Symptom	What to do
Tools missing / 401	Re-run the sign-in step (claude mcp login motionvid, codex mcp login motionvid, or reconnect the connector).
Forbidden / upgrade message	The account needs a paid plan: https://motionvid.ai/pricing
Not enough credits	Tell the user the balance from get_credit_balance; suggest a shorter video or a top-up.
failed with "timed out"	The scenes it built are saved: edit_video with "finish the remaining scenes".
429 / rate limited	Wait a minute, then retry once.

References

• Agents overview (markdown): https://motionvid.ai/agents.md
• MCP setup per client: https://motionvid.ai/mcp
• REST API: https://motionvid.ai/api-docs
• Index: https://motionvid.ai/llms.txt