vmeg_*). Unless noted, do not pass projId — the authenticated OAuth session or API Key supplies the project.
Field-level JSON schemas are returned by the MCP
tools/list response. This page summarizes names and behavior. REST equivalents (where they exist) live in API reference.Materials
| Tool | Description |
|---|---|
vmeg_list_materials | List uploaded video/audio materials in your project. Returns materialId, name, category, URLs, and metadata. Supports filters (category, keyword, materialId, mimeType) and pagination. |
vmeg_delete_material | Delete an uploaded material by materialId. Requires explicit user confirmation before the agent calls this tool. |
POST {mcpMaterialUploadUrl} with multipart/form-data file field. Returns materialId for create-task tools.
Voices
| Tool | Description |
|---|---|
vmeg_list_basic_voices | List system (library) voices. Filter by locale (BCP-47, e.g. en-US). Use with manual voice selection. |
vmeg_list_clone_voices | List your cloned voices (voiceId, provider, preview URLs). Use with voiceGenerationMode=manual. |
Tasks — create and manage
| Tool | Description |
|---|---|
vmeg_create_video_translation_task | Create a video or audio translation task. Requires targetLanguage and one media source: materialId, sourceUrl (HTTPS direct link, e.g. ChatGPT attachment), or youtubeUrl (video only). Product is auto-detected from material category. Ask the user for voiceGenerationMode (clone, clone-accent default, smart, manual). Returns taskId, status, product, editorUrl. |
vmeg_create_subtitle_translation_task | Create a subtitle translation task. Modes: preview with media + targetLanguage; merge with subtitleType; render/export with taskId + translatedScripts. YouTube imports may return pending while download runs asynchronously. |
vmeg_list_tasks | List or poll tasks (lite mode, no heavy config JSON). |
vmeg_delete_task | Delete a task by taskId. Requires explicit user confirmation before calling. |
Tasks — editor queries
All query tools taketaskId and optional filters. Times are on the post-translation timeline unless noted.
| Tool | Description |
|---|---|
vmeg_query_task_scripts | Query translated script lines (content.translated_scripts). Filter by keywords, time range, index range, speakerId. |
vmeg_query_task_subtitles | Query subtitle clips (content.subtitle_clips). |
vmeg_query_task_asr | Query ASR lines for transcription tasks (content.asr_scripts). |
vmeg_query_task_speakers | Query speakers (content.voice_match, or distinct script speakers). |
vmeg_query_task_tracks | Query video/audio track blocks (content.build_tracks). Filter by kind (audio / video). |
Tasks — editor mutations
| Tool | Description |
|---|---|
vmeg_save_task_draft | Apply draft edits in one transactional call. Pass ops: an array of typed operations (entity, action, value, …). All-or-nothing. |
vmeg_retranslate_task_scripts | Re-translate all scripts for an existing task. Confirm with the user before calling. Poll with vmeg_get_retranslate_status. |
vmeg_trigger_video_translation_all_tts | Re-dub all (or selected) translated scripts. Confirm with the user before calling. Poll with vmeg_get_tts_status. |
vmeg_compose_task_draft | Render (compose) the saved draft into a paid export job. Confirm with the user before calling. Returns a new render taskId. |
vmeg_get_retranslate_status | Poll whole-task re-translation status. |
vmeg_get_tts_status | Poll per-script TTS dubbing status. |
Agent behavior (server instructions)
The MCP server injects instructions for connected clients. Highlights:- Never pass
projIdin tool arguments. - Do not call create-task tools without a media source — ask how the user wants to provide media first.
- Confirm before delete, compose, re-translate, or bulk TTS.
- Tasks are asynchronous; poll
vmeg_list_tasksand editor query tools until work completes.
OAuth challenge
If a tool is called without a valid Bearer token, the server returns an MCP error with_meta OAuth challenge headers so clients (e.g. ChatGPT Apps SDK) can start login. After authorization, retry the tool call.