Short answerThe MCP server exposes the pipeline itself — runs, the bible, episode scripts, the world pack, art and renders, shot fixes, covers, sign-off, the delivery package, clips, Auto Sound, scheduled posts and your credit balance — as tools that map one-to-one onto what the console does. Approval gates and watching the episodes stay yours; the agent hands you a console link at each gate.
Each tool is a thin proxy for the same server-side action the console triggers; validation, gates and pricing are identical.
| Stage | Tools | Notes |
|---|---|---|
| Runs | list_runs, get_run, create_run, delete_run | create_run needs a premise (100–6,000 characters), a title in the primary language and compliance_ack: true. Episode count is fixed at 10 for self-serve accounts. |
| Bible (step 2) | generate_bible, get_bible, update_bible, sync_bible, approve_bible | generate_bible accepts feedback for a rewrite. get_bible reports stale_eps when beats lag behind the scripts. |
| Scripts (step 3) | generate_scripts, get_script, update_script, approve_script | generate_scripts resumes by default and skips finished episodes; force: true rewrites the whole chain. Episode numbers look like 0001. |
| World pack (step 4) | generate_pack, update_pack, commit_pack | Committing the pack creates the episodes and opens art and rendering. |
| Art & render (step 5) | art_estimate, generate_art, art_status | generate_art spends credits; run art_estimate first and confirm with the user. |
| Covers | covers_estimate, generate_covers, set_cover_theme, regen_cover | Cover generation is priced per image; the estimate lists which episodes lack one. |
| Finishing | signoff_episode, build_delivery_pack | Sign-off is a human decision: the agent should only call it after you have watched the episode. The delivery package is built only when you ask. |
| Account | credits_balance, get_review_url | get_review_url makes no API call; it returns the console deep link for the run. |
| Review & rework | render_status, update_shot, render_shot, rerender_group, retake_range, propose_fix, adopt_take, split_shot, insert_beats, grab_frame, update_staging, set_scene and the rest of the console agent's tools | These come from the console's built-in agent when the MCP server starts, so they follow the same review rules as the console. Fix the text first, then re-render only the shot that needs it. Tools that spend credits need user_message (your latest words, verbatim). After two attempts on the same shot the server stops and waits for a new message from you. |
| Clips | upload_video, clips_create, clips_get_job, clips_edit_clip, clips_send_to_matrix, clips_list_jobs, clips_series_episodes, clips_retry | Turns any video people talk in, or episodes of your own series, into vertical clips with pop captions and an end card. Charged 1 cr per started source minute (3 cr minimum) once the source is ready, refunded if no clip comes out. Poll with clips_get_job every 10 seconds or more. |
| Auto Sound | upload_video, auto_sound, auto_sound_status, auto_sound_export | Adds missing sound effects, an ambience bed and ducked music to an uploaded video, at Subtle, Standard or Punchy intensity with effects, ambience and music as separate layers. Charged 5 cr per started minute of footage the project has not processed before (1 minute minimum, up to 15 minutes per pass), refunded if the job fails. Poll with auto_sound_status every 10 seconds or more, then auto_sound_export renders the finished video. |
| Drama Center | dramacenter_check, dramacenter_get, dramacenter_update, dramacenter_package, dramacenter_delete, dramacenter_list | Checks a season against TikTok Drama Center's review rules (dark edges, web text, subtitle area, cover frame, metadata), builds the upload package, and deletes old checks. The light check is free; the full check, fixes and package cost 2 cr per episode. |
| Publishing (posting) | matrix_accounts, matrix_update_account, matrix_lanes, matrix_update_lane, matrix_media, matrix_add_media, matrix_plan_posts, matrix_schedule_posts, matrix_posts, matrix_cancel_post, matrix_comments, matrix_report, matrix_status | Schedules clips across your connected accounts. Always run matrix_plan_posts before matrix_schedule_posts, and pass a client_token so a retry never double-posts. Matrix is in private beta. |
Two tools, import_clip and upload_character_ref, are reserved for Dawnwood staff and return an error on customer accounts.
Three actions change what the pipeline is allowed to do next: approve_bible, approve_script and commit_pack. Well-behaved agents call get_review_url before each of them and stop. Open the link, read the bible or script in the console, and either approve there or tell the agent to approve. Rendering starts only after the pack is committed, so nothing burns credits that a human did not clear. Sign-off works the same way: the agent may call signoff_episode, but only after you have seen the episode master.
If you ask the agent to approve without reviewing, it will comply — the gate is a workflow convention, not a lock. Keep the review step if you want the safety.
Bible generation, the script chain, art and covers run in the background. The agent tracks them by polling:
get_run for the wizard steps, art_status for art and rendering, covers_estimate for covers.list_runs is for browsing, not polling. It is expensive and returns every run.The tool descriptions carry these rules, so most agents follow them without being told.
The same price card as the console applies; see what costs credits. Over MCP the spending tools are generate_art (art and renders), generate_covers and regen_cover (covers), and script edits made after an episode is approved. Bible and script generation before approval are free, which means a Free account can take a premise through the full bible and every script without spending anything. Failed generations are not charged.
Every charge is tagged with the token that made it. API tokens in the console shows credits spent per token, and credits_balance returns a breakdown by source (console versus each agent name) and by token.
Every run, version and charge an agent creates appears in the console exactly as if you had clicked it. The bible and script pages keep version history with the actor recorded, the Credits page itemises spending by source, and Where is my history in the console answers the rest; see where is my history.
Still stuck? Email support@dawnwood.ai with your account email, the series, the episode and the segment or group — see what to include.