aicut

aicut MCP server

Generate AI story episodes, videos, images and audio straight from Claude, Claude Code, ChatGPT, Cursor, Grok or any MCP client - no API key, no dashboard round-trip.

• Server URL: https://mcp.aicut.pro/mcp
• Transport: Streamable HTTP
• Auth: OAuth 2.1 with dynamic client registration + PKCE. You sign in with your normal aicut account and approve once.
• Setup page: https://www.aicut.pro/mcp

This repository is the public metadata for the hosted server: there is nothing to install and nothing to run locally.

Install

Claude Code

claude mcp add --transport http aicut https://mcp.aicut.pro/mcp

Then run /mcp, pick aicut, choose Authenticate, and approve in the browser tab that opens.

Claude (web and desktop)

Settings -> Customize -> Connectors -> Add -> Add custom connector, name it aicut, paste https://mcp.aicut.pro/mcp, hit Connect and approve.

ChatGPT

Settings -> Security and login -> Advanced security -> switch on Developer mode. Then open Plugins in the sidebar, hit +, name it aicut, paste https://mcp.aicut.pro/mcp, leave Authentication on OAuth, tick the box and Create. Choose Sign in with aicut and approve.

Cursor

Customize -> MCPs -> New MCP Server, name it aicut, paste https://mcp.aicut.pro/mcp. Cursor lists it under Needs attention: hit Authenticate and approve. (Or use the one-click Add to Cursor button on https://www.aicut.pro/mcp.)

Grok

grok.com/connectors -> New Connector -> Custom, name it aicut, paste https://mcp.aicut.pro/mcp. The aicut sign-in opens: approve it. Custom connectors come with Grok's Business and Enterprise plans.

Any other MCP client

Point it at https://mcp.aicut.pro/mcp over Streamable HTTP. Clients that implement the MCP OAuth flow register themselves; clients with no browser can use an aicut API key as a bearer token instead (keys are issued by hand - email [email protected]).

Tools

61 tools, plus 12 MCP prompts (step-by-step recipes; get_recipe serves the same text to clients that do not show prompts). Any argument a tool does not declare is refused with an error that names it, rather than silently dropped. Read tools are annotated readOnlyHint; the tools that spend carry destructiveHint and idempotentHint: false, so a client asks before spending. Every generate_* tool takes estimate_only: true to price a request without creating anything - and on audio that is the only reliable quote, because audio models publish a per-character or per-second RATE rather than a price table. Generations spend tokens from the connected aicut account at the same prices as the app, and land in that account's library.

Catalog and account

Tool	What it does
list_models	The curated video, image and audio catalog with the exact cost of every settings combination. Call this first - model ids and their accepted values cannot be guessed. Searchable and paged.
get_balance	Tokens left on the account, and its plan tier.
list_voices	The whole voice catalogue - ElevenLabs (stock, plus this account's cloned and custom voices) and OpenAI in one shape, each with a verdict on what it can be used for. Paged.
list_characters	The account's uploaded characters and generated story cast members, in one list.
list_series	The AI Video Story catalog: which series (niches) an episode can be made in, with their episode-length ladder, model defaults and prices.
list_series_ideas	One series' curated episode ideas.
get_recipe	A step-by-step recipe for a flow, read before its first call: a first video, image or audio, an AI Video Story episode, a fake-text or image story script, a niche run, a talking avatar, motion control.
browse_series	Puts the series catalog on screen as a grid of covers whose demo clips play in place, on clients that render cards. Tapping selects, one Confirm writes the pick.

Video, image and audio

Tool	What it does
generate_video	Starts a video generation. Asynchronous: returns a job id.
get_video	One video by id, with its status and finished url.
list_videos	The account's recent videos, paginated.
delete_video	Deletes one of the caller's own videos. No refund.
generate_image	Generates an image, and usually returns it finished in the same call.
get_image	One image by id.
list_images	Recent images, paginated.
generate_audio	Speech, music or a sound effect, chosen by model.
get_audio	One audio job by id.
list_audio	Recent audio, paginated.
wait_for_generation	The long poll: waits server-side and answers whether a job is terminal.
show_generation	Replays an earlier generation the user asks to see again.

Transforms - working from media you already have

What the server measures IS the price: a clip's length, a picture's dimensions, an audio's length. Media the server cannot measure is refused before anything is charged rather than charged and then failed.

Tool	What it does
upscale_video	2x or 4x an existing video.
upscale_image	More pixels in an existing image, from three upscalers: a general-purpose one, one that goes to 10x and is priced per output megapixel, and a creative one that repaints detail as it enlarges. Priced from the picture's measured dimensions.
extend_video	Continues an existing video by a chosen number of seconds.
reshot_video	A 4-30 second clip plus reference pictures, remade either as an object swap (one subject or object replaced, everything else kept) or as a motion transfer (the reference subject performs the clip's moves). Needs the likeness consent accepted once in the aicut web app before the first use. Not offered on ChatGPT.
list_reshot_templates	The Reshot gallery: ready-made clips, each with a preview of its remake, any of which reshot_video takes as a template. Not offered on ChatGPT.
motion_control	A character picture plus a reference clip, and the character performs that clip's motion. The reference clip's length is the price.
generate_lipsync	A character picture plus audio generated on this account, and the character speaks or sings it, lip-synced. The result is exactly as long as the audio, and the audio's measured length is the price.
edit_video	A 3-10 second clip plus a description of what to change, and the SAME clip comes back with those changes - different cast, clothes or setting - while the camera move, timing and motion stay exactly as they were. Real identifiable people in the source can be refused by the provider, and a refused job is refunded.

Video analysis

Tool	What it does
analyze_video	Reads an existing video instead of making one. Spends no tokens; capped per account per day.
get_analysis	One analysis by id.
list_analyses	Recent analyses, paginated.
image_to_prompt	Writes a prompt that regenerates a similar image, from an image url. Spends no tokens; capped per account per day.

AI Video Story - the staged episode flow

An episode is made in stages, and each stage has its own price, so nothing large is bought in one blind step: the cast is written free and its portraits are bought; the episode's start frames are charged and PARK for review; firing buys the scene videos; rendering buys the final file.

Tool	What it does
generate_cast	FREE. Writes a story cast for a series and returns it to read. Mints nothing.
generate_cast_portraits	Buys one image per cast member and saves the roster to the library.
regenerate_cast_portrait	Redraws one cast member's face.
describe_cast_member	Changes what one cast member is, then redraws them.
generate_story_video	Stage 1: writes the episode, generates the start frames, and parks for review. estimate_only quotes the whole episode with its stage split.
regenerate_story_frame	Redraws one parked start frame.
change_story_scene	Changes what happens in one parked scene, then redraws its frame.
remove_story_scene	FREE. Removes a parked scene, so its video is never generated and never charged. Replaces set_scene_kept, which still answers as an alias.
fire_story_video	Stage 2: generates the scene videos for the scenes you kept. Irreversible, and an episode cannot fire twice.
render_story_video	Stage 3: the final render. Free-tier renders carry a watermark.

One-call formats

These write and render in a single call: no fire step, no render step.

Tool	What it does
generate_fake_text_video	The fake-text chat-story format. The agent writes the message script; the server speaks it, measures the bubbles and renders it. The voice provider moves the price, so re-quote after changing it.
generate_image_story	The AI image story format. The agent writes the narration; the server segments it, writes the image prompts, generates every scene image, speaks it and renders it. Price is driven by the image count.
list_niches	aicut's one-shot trending formats ("niches"): each made from a few answers, and sometimes a picture, at one price. Without a slug, the cards; with one, that niche in full.
generate_from_niche	Runs one niche on the account and returns the job id at once; the video is collected with wait_for_generation.
list_image_story_styles	aicut's authored image-story looks. Optional on a create, and picking one forces the pricier edit-capable image model - re-quote after adding one.

Publishing to social

Tool	What it does
import_video	Free. Brings a clip not made on aicut - the user's own upload or a public link to an MP4 or MOV - onto the account as a finished video, so it can be posted.
list_social_accounts	The TikTok, YouTube and Instagram accounts this user has connected. An empty list means they must connect one in a browser - this API cannot.
prepare_post	Stages a post of one of the account's finished videos, named by aicut id. Publishes nothing, and expires after 30 minutes.
review_post	The consent card: shows the actual video, the destination account and the caption, visibility and disclosure controls, and re-stages the person's own choices. Publishes nothing.
publish_post	IRREVERSIBLE: sends or schedules the prepared post. States the per-account rate ceiling rather than making an agent discover it.
get_post_status	What the platform says happened, from prepared through to posted or failed.

Trends

Tool	What it does
list_trends	aicut's published trend feed. Read-only.
get_recreate_brief	How to recreate one trending video. Names the niche it can prove, and deliberately names no series.

Uploads - using your own footage and images

Tool	What it does
upload_media_widget	The in-card file picker. The person chooses a file, the card uploads and confirms it, and posts the finished url back. The agent's whole job is to call this and wait.
upload_media	A signed upload target for a file on the user's own machine, plus the curl that fills it.
list_uploads	Free. The account's recent uploads, last used first, each with its url and kind.
show_uploads	Free. The same list opened as a numbered grid of thumbnails the person picks from in the chat.
import_link	Free. A TikTok link becomes a 3-30 second clip on the account, usable as any video input.
confirm_upload	The second half: re-reads the uploaded object, checks its magic bytes against the declared type and measures a video, refusing anything that fails.

Docs and policies

• Setup and FAQ: https://www.aicut.pro/mcp
• Privacy policy: https://www.aicut.pro/privacy
• Terms: https://www.aicut.pro/terms
• Support: [email protected]