Yaps Auto Captions

Yaps Auto Captions

Use the existing Yaps MCP tools. Do not invent CLI commands or use a hosted captioning service as a fallback.

1. Check the exact source with captions_verify and inspect templates with captions_styles.
2. Create an editable project with captions_create. Use the returned project ID for subsequent actions.
3. Inspect captions_show. Apply requested edits with captions_correct, captions_replace, captions_split, captions_merge, or captions_set_style. Preserve the spoken wording unless the user asks for a correction. captions_reset discards edits, so use it only after explicit approval.
4. Render with captions_render to a separate .mp4, then verify the result with captions_verify. Never target the source. Only set overwrite: true after approval to replace the exact existing output.

Return a link to the captioned video. A saved caption project is not a completed video export. Use the transcribe skill for plain text and srt-generator for a separate .srt file.

Reachability

Call yaps_status when a tool reports a problem. If it returns local_yaps_unreachable, cli_missing, or equivalent, the current session cannot see the Yaps engine. Do not claim Yaps is uninstalled. Offer Download Yaps, ask the user to open Yaps, and retry from a local session on the same computer. Only use yaps_enable_feature to install a model after the user requests it.

Standalone plugin scope

This plugin supplies only the Yaps Auto Captions workflow. Other Yaps skills mentioned above are separate plugins. Use another workflow only if its tools are actually installed; otherwise explain which plugin is needed. Do not invent missing tool calls.

Check yaps_status for feature and model readiness when setup is missing. If a required model is absent, explain the download and ask once unless the user already authorized that installation. Use yaps_enable_feature with an allowed feature and appropriate engine, check readiness again, then resume the original task. Reuse existing models; do not delete or reinstall working models.