Moknah

Credits for a whole PDF book 'as imported' (OCR + optional normalize/ translate + full TTS). char_count is approximated from pages if omitted.

Credits to (re)render only the edited/added lines of a project.

Credits to render a project. scope='all' or 'dirty' (changed lines only).

Credits to synthesize a piece of text with an optional specific voice.

Render TTS audio for a whole project or one chapter. Provide exactly one of project_id or chapter_id. mode='missing' renders only lines without audio yet (use after edits); mode='all' regenerates everything. BILLABLE: this starts a credit-consuming job. Bills exactly like the web app (resolved voice settings + per-character rate + premium-voice fee) and checks the balance first. Call estimate_project_generation beforehand to preview the cost. Returns a job_ref to poll with get_job; fails clearly if no voice is set.

Render audio for one or more specific lines, by line id. Use this for "generate line X" or a handful of lines; use generate_audio for whole chapters/projects. Runs in the background (TTS is too slow to block on) and returns a job_ref -- poll get_job(job_ref) until is_terminal, then the rendered line ids are in the result. BILLABLE: resolves each line's voice cascade and bills like the web app. A no-voice / insufficient-credits problem surfaces as a failed job with a clear message; estimate_edits previews the cost.

Return the caller's current credit balance.

Get one chapter with its lines (text, QA status, per-line voice, audio state). Bounded to a single chapter (<=200 lines) so the payload is safe. Use after get_project shows you the chapter list, to read line-level content.

Get the merged audio file for a whole chapter (the single stitched MP3). Returns the public Spaces URL of the chapter's current merged version, or has_audio=false if it hasn't been rendered/merged yet. Produce it with generate_audio(chapter_id=...) (poll the job); this just fetches the result.

Poll a long-running job by its job_ref. is_terminal=true when finished; status is one of pending/processing/completed/failed/cancelled.

Get a finished job's output as a public URL (audio / transcript / file). This does NOT download bytes -- it returns the URL to fetch. Poll get_job until is_terminal, then call this for the artifact.

Get the rendered audio URL(s) for one or more lines, by line id. Returns each line's current audio version: audio_url, the voice it was rendered with, and the version id. Lines not yet rendered come back with has_audio=false. Use this after generate_lines finishes (its result lists the rendered_lines), or for any already-rendered line.

Project overview: metadata + a per-chapter summary (line counts and char totals) WITHOUT line text, so it stays small even on multi-million-character books. To read the actual lines of a chapter, call get_chapter(chapter_id).

Read AI word-level QA results for a chapter (produced by run_ai_qa). Returns the QA run status and, per line that has issues, the suggested corrected text and the list of pending issues; plus totals so you know if review is still needed. status='none' means run_ai_qa hasn't been run yet.

List the caller's audiobook projects (most recent first). Page with `offset`; the result includes total and has_more for large accounts.

List voices available to the caller (owned, team, or free public). Page with `offset`; the result includes total and has_more.

Merge rendered chapters' audio into ONE MP3 (book order) and return a public URL. FREE -- pure audio concatenation, no TTS. Same feature as the web's Merge button. Chapters without rendered audio are skipped; needs at least 2 with audio. Use for the final audiobook file or combining an intro with its first chapter.

Merge chapters' word-level subtitles into ONE SRT or VTT file with continuous timestamps (book order) and return a public URL. FREE. Same feature as the web's subtitle merge. Requires rendered chapters (word alignment is produced during generation). Pairs with merge_chapters_audio to deliver a single audiobook file + matching subtitles.

Show an inline audio PLAYER for Moknah audio (MCP Apps widget). Pass a Moknah audio_url (or a finished job_ref) and this renders the ui://moknah/audio-player widget -- a real play button in hosts that support MCP Apps (e.g., ChatGPT). Hosts without MCP Apps get the audio_url in the result instead. SECURITY: only audio hosted on Moknah (moknah.io / *.moknah.io) is allowed; any other URL is refused, so the player can never be pointed at an external or untrusted source.

Closure test: return a line's rendered audio as an MCP `audio` content block (base64) plus the URL as text. Open the raw tools/call result in MCP Inspector to confirm the audio block is delivered; if it is but Claude shows no player, that's Claude's renderer, not the server. (One short line only.)

Show an inline audio PLAYER for a voice's sample clip so the user can hear it before choosing. Returns structuredContent (title, subtitle, audio_url) and links the ui://moknah/audio-player widget, so hosts that support MCP Apps (e.g. ChatGPT) render a real play button in chat. The audio_url is the voice's public sample URL. NOTE: Claude Desktop/claude.ai don't render MCP Apps yet (upstream) -- they get the audio_url as text. Errors if the voice has no sample or isn't available to the caller.

Rename a chapter (does not touch its lines or audio).

Reorder a project's chapters to match the given list of chapter ids.

Reorder a chapter's lines to match the given list of line ids.

Run AI word-level quality checking over a chapter (async job). BILLABLE: the AI QA pass consumes credits. Returns a job_ref -- poll with get_job, then read findings with get_qa_results.

Set a line's QA status (1=NotStarted ... 6=Finalized).

Override voice settings for one line (voice_id, temperature, similarity, speed, expressiveness, emotions_mode, prerecording). Null = inherit.

Set the default voice for a whole project.

Generate speech from raw text -- no project needed. Available on EVERY plan, including free tier (Audio Studio itself is a Pro feature). BILLABLE when confirm_spend=true (per character + premium-voice fee). The default confirm_spend=false call is FREE and returns only the estimate -- show it to the user and get their OK first. With confirm_spend=true the render runs in the BACKGROUND (TTS is too slow to block on) and returns a job_ref: poll get_job(job_ref) until is_terminal, then get_job_result(job_ref) for the public audio_url.

Transcribe an audio file to text (speech-to-text). Provide the audio as base64; `filename` gives the type (mp3/wav/m4a/aac/ogg). BILLABLE: bills `stt_api` per minute, estimated from the audio's real duration, and refuses if credits are short. Runs in the background and returns a job_ref -- poll get_job(job_ref) until is_terminal; download_result then returns the transcript file's public URL.

Edit a line's text. Marks the line unconverted so it re-renders.

Batch-edit many lines in a single request: text, voice settings, and/or QA status per line. FREE (no credits). Items are applied independently -- the response reports per-item success/failure, so one bad line does not abort the rest. Prefer this over calling update_line/set_line_voice_settings repeatedly.