Tool reference

Screel serves its tool catalogue on five MCP surfaces. Which tools appear, and which shapes they take, depends on where you connect:

Surface	Endpoint	Tools
ChatGPT app	https://api.screel.ai/mcp/chatgpt	24 — management + generation, with in-chat widgets
Second ChatGPT app	https://api.screel.ai/mcp/chatgpt-v2	26 — a separate app, submitted independently, with a merged tool set, reference documents, and its own in-chat widgets
Claude connector	https://api.screel.ai/mcp/claude	17 — management and rendering only (no AI media generation)
Muse connector	https://api.screel.ai/mcp/muse	26, listed in the Muse guide — the second ChatGPT app’s tool set, with the same names and widgets; the tools that take a chat file attachment there take a public https image URL here instead (image_url, reference_image_urls)
Generic MCP	https://api.screel.ai/mcp	24 — management + generation, plain JSON responses

Two OAuth scopes gate the tools: screel.read for read-only tools and screel.generate for anything that creates or changes content. Tools marked free never spend credits. On the second ChatGPT app, every credit-spending tool also takes an optional request_id: send the same one when you retry a call that may already have started, and the result comes back without spending again.

Account and browsing

Available on every surface, except where noted.

Tool	What it does
get_account	Your credit balance, storage usage, enabled capabilities, and the per-action credit costs.
list_projects	List your video projects, newest first, with thumbnails and scene counts.
get_project	A project’s full storyboard: every scene with its image, clip, voiceover and status, plus project-level audio. A project linked to an Unreal sequence also reports linked and, per scene, dirty_reasons — why a bound scene’s clip no longer matches its marker — plus binding.in_frame, the objects the capture reports with the reference document each one links to, and binding.shot_size. On the second ChatGPT app, it also reports the project’s reference documents and render settings, and takes a detail of full or summary for a lighter progress check.
get_scene	One scene’s full details, addressed by ID or position. Not on the second ChatGPT app — read it from get_project there.
list_renders	All renders of a project with their status and video URLs. Not on the second ChatGPT app — pass all to get_render there instead.
get_render	One render (or a project’s latest) with its playable video URL, poster and file details. On the second ChatGPT app, pass all with a project ID to list every render of that project instead of one.
get_job_status	Progress of up to 40 generation/render jobs in one call, including queue position for waiting jobs and result URLs when done.

Projects and scenes

Available on every surface, except where noted.

Tool	What it does
create_project	Start a new project with a name, optional description, and aspect ratio (16:9, 9:16, 1:1, 4:3).
add_scenes	Append up to 20 scenes in one call — the whole script at once, each with a name, description, and an optional image prompt and duration.
update_scene	Update a scene’s settings: name, description, position, duration, caption text, caption visibility, audio generation, voiceover style and voice. Not on the second ChatGPT app — use update_scenes there.
update_scenes	Second ChatGPT app only. Batched scene edits in one call: update settings on several scenes, reorder, duplicate or delete them — replacing update_scene, duplicate_scene and delete_scene on that surface.
update_project	Change a project’s name, description, aspect ratio or background-music volume. On the second ChatGPT app, also change its render settings — resolution, frame rate, quality, transitions, fades, captions, color grade, audio and the motion on scenes that render as stills; unrecognized values are rejected, and a setting the account can’t fully honor comes back with a one-line warning.
duplicate_scene	Copy a scene; the copy lands right after the original. Not on the second ChatGPT app — use update_scenes there.
delete_scene	Remove a scene. Destructive — hosts ask for confirmation. Not on the second ChatGPT app — use update_scenes there.
set_scene_image	Attach an existing image to a scene as its still — a chat file attachment in ChatGPT, a public image URL elsewhere. On the second ChatGPT app it also restores a version from the scene’s history in place of revert_scene_image. Free.
revert_scene_image	Restore a scene’s previous image or clip version from its history. Not on the second ChatGPT app — pass a history version to set_scene_image there. Free.

Rendering

Available on every surface.

Tool	What it does
render_project	Compose the project’s scenes — stills, clips, voiceover, music — into a finished MP4 (720p–2160p; 1440p and above fall back to 1080p when the account does not have that capability). Format and resolution default to the project’s render settings. Scenes without a clip render as stills. Free.
cancel_job	Cancel one or many queued or running jobs; cancelling a waiting job is free.

Generation

Available in the ChatGPT app, the second ChatGPT app, and the generic MCP endpoint. These spend credits. They return a job_id immediately; progress arrives via get_job_status.

Tool	What it does
generate_scene_image	Generate a scene’s image from a text prompt — one scene or a batch of up to 20 in a single call. Not on the second ChatGPT app — use generate_image there.
refine_scene_image	Edit a scene’s existing image with an instruction, optionally guided by up to 14 reference images. Not on the second ChatGPT app — use refine_image there.
generate_image	Second ChatGPT app and generic MCP. Generate an image from a text prompt — a scene’s still, a picture for a reference document, or a standalone image, one at a time or in a batch of up to 20 that can mix all three. Guided by up to 14 reference images and, on the second ChatGPT app, by reference_document_ids naming reference documents a generation must match, so a recurring character, object or place stays consistent across scenes; the references are written into the scene’s prompt as mention tokens, so they show in the editor and survive a regenerate. On the second ChatGPT app it also takes chat file attachments as reference_images. A batch item may name its own reference_image_urls (for example the previous scene’s still, for continuity), and a picture that lands in a document takes an image_name such as front or side.
refine_image	Second ChatGPT app and generic MCP. Edit an existing image with an instruction — a scene’s still, a document’s picture, or a standalone image — one at a time or batched, with the same reference-image and (on the second ChatGPT app) reference_document_ids support as generate_image.
animate_scene	Turn a scene’s image into a moving clip from a motion prompt (3–10 s) — batchable. resolution picks the tier (360p draft by default on the ChatGPT app; 720p, 1080p, 2160p); on the ChatGPT app end_frame (next, a scene id or a Screel image url) makes the clip end on that picture so consecutive scenes flow into each other.
generate_voiceover	Text-to-speech narration for a scene, with selectable voice, style and speed — batchable.
generate_music	Generate a background music track for the project from a prompt. The track length depends on the account — get_account reports it.
upscale_render	Upscale a finished render to 1080p, 1440p or 2160p. Not available on every account — get_account reports whether it is.
list_voices	The available voiceover voices and style presets, each with a one-sentence description, plus the defaults. Free.

Reference documents

Available only on the second ChatGPT app. A reference document holds a title, markdown notes, and a small picture bank; reference_document_ids on generate_image and refine_image names which documents a generation must match.

Tool	What it does
list_documents	List a project’s reference documents, each with the folder it sits in, plus the project’s folders.
get_document	One document’s content and picture bank.
create_document	Start a reference document with a title and markdown blocks.
edit_document	Replace a document’s blocks.
set_document_image	Add or replace a picture in a document’s bank — a chat file attachment or the URL of a picture Screel already stores.
get_playbook	One of Screel’s playbooks for reference work — consistent-characters, props-and-objects, references-from-uploads, scene-continuity or style-and-world — as text. The same playbooks ship as skills with the second ChatGPT app. Free.
open_project	A link that opens the project in the Screel web app, for work this surface has no tool for: fine-tuning scene timing, editing a document in place, browsing or restoring media, sharing, comments, or previewing voices. Costs no credits and changes nothing. Free.

Document folders

Available only in the in-app agent (Manan), not on any MCP surface. A project’s documents sit in folders in the editor sidebar; a document never nests inside another document. list_documents reports the folders and the folder each document sits in.

Tool	What it does
create_folder	Add a folder to the project’s document sidebar, optionally inside another folder and with a colour. Free.
move_document	Move a document into one of the project’s folders, or back to the sidebar root. Free.

Whiteboard

Available only in the in-app agent (Manan), not on any MCP surface.

Tool	What it does
get_board	A whiteboard’s items — pictures, notes, text, groups and arrows — by group and reading order, with each item’s grid cell.
edit_board	Batched whiteboard edits: add a picture, note or text item (add_text, with a size and alignment), link a document as a card (add_document, with a document_id), label, tag, caption, group, arrange, connect or remove items, or set the cover. move and remove refuse a locked item.

Breakdown

Available only in the in-app agent (Manan), not on any MCP surface. A breakdown document reads one picture and lists the assets in it as rows, each with a category, a description and, once generated, a reference picture. A mention of a breakdown in the chat reaches Manan as a summary that points at get_breakdown; the rows and their pictures come from the tools.

Tool	What it does
get_breakdown	The sheet: source picture, categories, notes and every asset row with its status, box and reference picture URL.
edit_breakdown	Batched row edits: set, add or remove an asset, accept or reject proposed rows, set the notes or the category list, reorder.
breakdown_image	Attach a scene still, an upload or a document picture as the source and run the vision pass that proposes asset rows. Creates the document when none is named.
generate_asset_references	One reference picture per accepted asset without one, cut from the source around the asset’s box. Charged at the image rate.

Live engine session

Available only in the in-app agent (Manan), not on any MCP surface. Both tools work only on a project linked to an external sequence, and only while a session for it is connected.

Tool	What it does
host_capture	Tell the connected engine session to capture one or more Sequencer markers back into their scenes’ stills — a named selection or every bound scene.
host_seek	Move the connected engine session’s playhead to the Sequencer marker one scene is bound to.