Speechdub

View saved version →

---
name: browse-speechdub-library
description: List or open documents in the user's Speechdub library when they want to see saved titles, search by name, or read full document text.
---

# Browse Speechdub library

Use this skill when the user asks to see, search, or read documents they already saved in Speechdub (not when they only want new text pasted in chat).

## Tools

- `speechdub_list_documents` — paginated list; optional `query` for title search, `limit` (max 100), `cursor`, `include_archived`.
- `speechdub_get_document` — full title, language, and body for one `document_id` (UUID).

## Workflow

1. Prefer `speechdub_list_documents` when the user does not give an id. Use a reasonable `limit` (e.g. 10–20) unless they ask for more.
2. If they name a document, pass `query` on list, then `speechdub_get_document` on the best match.
3. If they provide a UUID, call `speechdub_get_document` directly.
4. Summarize titles and languages clearly; when they ask for content, quote or summarize the returned text without inventing library items.

## Boundaries

- Read-only: do not create, update, delete, or synthesize unless the user clearly asks for those actions (use other skills).
- Never expose API keys or raw OAuth tokens. The MCP connection is already authenticated as the user.

View saved version →

---
name: edit-speechdub-document
description: Rename or update text and language of an existing Speechdub document the user owns, or permanently delete a document when they explicitly request removal.
---

# Edit or remove Speechdub documents

Use when the user wants to **change** a saved document (title, body, language) or **delete** one they identify by name or id.

## Tools

- `speechdub_list_documents` / `speechdub_get_document` — resolve `document_id` when the user only gives a title or vague reference.
- `speechdub_update_document` — `document_id` plus any of `title`, `text`, `language`.
- `speechdub_delete_document` — `document_id` only; **permanent**.

## Workflow

1. Resolve the target document id (list + get if needed). Do not guess UUIDs.
2. For renames or content edits, use `speechdub_update_document` with only the fields that should change.
3. For deletion, confirm the user intent when ambiguous (title + id). Then call `speechdub_delete_document` once.
4. After delete, state clearly that the document was removed and cannot be restored via the API.

## Boundaries

- `speechdub_delete_document` is destructive; do not use it for “archive” or “clear chat” requests unless they mean delete from Speechdub.
- Do not create new documents here (use import-to-speechdub).

View saved version →

---
name: import-to-speechdub
description: Save new text into the user's Speechdub document library from pasted content, notes, or articles they want to listen to later in the app.
---

# Import text into Speechdub

Use when the user wants to **add** content to their Speechdub library (import, save, create a document), not merely read it in chat.

## Tool

- `speechdub_create_document` — required `text`; optional `title`, `language` (BCP-47-ish code), `idempotency_key` for safe retries.

## Workflow

1. Collect the text to import. If the user attached or pasted content, use that as `text`.
2. Choose a short, descriptive `title` if they did not specify one (plain text, no markdown formatting in the title).
3. Set `language` when the user specifies it or when the source language is obvious; otherwise omit and let Speechdub detect when possible.
4. Call `speechdub_create_document` once per distinct document they asked for.
5. Confirm success with returned `document_id`, title, and language. Mention that creating documents uses their Speechdub document quota (not the same as speech credits).

## Boundaries

- Do not delete or overwrite existing documents in this skill.
- Do not call speech synthesis unless they also ask to hear audio (use the text-to-speech skill).

View saved version →

---
name: speechdub-text-to-speech
description: Generate spoken audio with Speechdub preset voices when the user wants text read aloud, a voice preview, or audio output from their content.
---

# Speechdub text-to-speech

Use when the user wants **audio** or **read aloud** via Speechdub voices, not only silent text in chat.

## Tools

- `speechdub_list_voices` — optional `gender` filter (`male` | `female`).
- `speechdub_get_voice` — metadata for one `voice_id` (presets like `F1`, `M3`).
- `speechdub_synthesize_speech` — required `input` (max 5000 chars) and `voice_id`; optional `language`, `audio_format` (`wav` | `mp3`), `include_audio_data`.

## Workflow

1. Always call `speechdub_list_voices` (or `speechdub_get_voice` if they named a preset) **before** synthesis unless a valid `voice_id` is already confirmed.
2. Pick a voice that matches the user’s language or preference; say which voice you chose.
3. Split long text into chunks under 5000 characters if needed; synthesize per chunk.
4. Report billing context briefly: synthesis uses the user’s Speechdub credit wallet (same as the web app). Do not quote API keys.
5. If `include_audio_data` is false or omitted, describe the result (duration, format, request id) per tool output; follow the client’s rules for playing or attaching audio.

## Boundaries

- Do not create or delete library documents unless the user also asked for library changes.
- Prefer synthesis on user-supplied or document text they explicitly want heard; do not synthesize huge documents without narrowing scope.