About longtake
longtake-mcp connects Longtake Studio’s video-production workflow to AI agents through MCP. It supports project and character setup, draft generation, 1080p finishing, and per-render cost checks.
Key features
- 01Manage projects, scenes, and characters
- 02Upload image, audio, and video assets
- 03Generate character sheets and video takes
- 04Estimate costs and check render status
- 05Finish Seedance 2.5 drafts in 1080p
Use cases
- 01Set up project and scene structures
- 02Give characters portraits and locked voice samples
- 03Generate low-cost drafts before final renders
- 04Review the cost of each video render
MCP support
The MCP server exposes Longtake Studio production tools to MCP clients, including project management, asset uploads, video generation, and render-status checks. Rendering uses the user's own model-provider API keys.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
npx -y longtake-mcpSTUDIO_URLStudio address — https://longtake.studio for the cloud studioSTUDIO_TOKENSecretPersonal agent token from the studio: account menu → API keys → Agent accessLocal servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.
Tools
13Saved tool list · updated 2w ago
This tool takes no parameters.
Details
From the project READMElongtake-mcp
MCP server for Longtake Studio — an AI video studio that runs on your own API key. With this server an AI agent (Claude Code, Codex, Cursor, any MCP client) can set up a production, cast characters with a locked voice, shoot takes on Seedance 2.0 / 2.5, Veo 3.1, Gemini Omni or Kling 3.0, draft cheaply and finish in 1080p, and read back the cost of every render.
Renders launched by an agent are billed exactly like clicks in the studio: the model provider (BytePlus, Google, Kling, fal.ai) charges your own key at its list price. Agent access itself is neither metered nor marked up.
Longtake is the name of the software. It is not the cinematography term "long take" — though that is where the name comes from: one take of one scene is the unit of work.
What the agent gets — 13 tools
| Tool | What it does |
|---|---|
studio_overview | Projects, episodes, scenes, characters and assets — the whole studio at a glance |
create_project · create_folder · create_scene | The production structure: project → episode → scene |
create_character | Cast a character with a portrait and a locked voice sample; the studio screen-tests the portrait against moderation |
generate_mocap | A four-view character sheet (Seedream / Nano Banana Pro) |
upload_asset | Reference images, audio and video clips for the project library |
estimate_cost | The price of a take before it is shot — model, resolution, length, reference clips |
generate_video | Shoot a take; @names in the prompt resolve to characters and assets; draft: true shoots a Seedance 2.5 draft |
finish_draft | Finish a Seedance 2.5 draft in 1080p — the same take, rendered sharper |
take_status · scene_takes | Status, file, token spend and what a take was shot as |
cancel_take | Delete a take (a queued BytePlus render is cancelled at the provider) |
Setup
-
Issue a token. In the studio: account menu → API keys → Agent access — MCP → create. It is shown once and can be revoked alone.
-
Register it with your agent. The package is on npm as
longtake-mcp, sonpxfetches it — nothing to clone. For Claude Code:claude mcp add longtake \ -e STUDIO_URL=https://longtake.studio \ -e STUDIO_TOKEN=cf_your_token \ -- npx -y longtake-mcpAny other MCP client takes the same command and the same two environment variables. As JSON (Cursor, Codex, Claude Desktop):
{ "mcpServers": { "longtake": { "command": "npx", "args": ["-y", "longtake-mcp"], "env": { "STUDIO_URL": "https://longtake.studio", "STUDIO_TOKEN": "cf_your_token" } } } }Prefer a checkout?
git clone https://github.com/svatrous/longtake-mcp.git && cd longtake-mcp && npm install, then point the command atnode mcp-server.js.
The full guide, with what the token can and cannot do, is at longtake.studio/mcp. Prompt-writing skills for the same agent (Seedance 2.0, Seedance 2.5, Gemini Omni, character sheets) are at longtake.studio/skills.
Safety
The token acts as your account for the work: it shoots takes on your provider keys, inside your rate limits. It cannot read or change those keys, issue or revoke tokens, or touch billing — those routes turn an agent token away. Everything else you can do it can do too, including deleting a project, so keep it out of shared configs and revoke it in the same dialog if it leaks.
Local files. Three tools read a file from this machine: upload_asset; create_character, when given a portrait or a voice sample by path; and generate_mocap, for its reference photos. They take only media — jpg, png, webp, gif, mp3, wav, m4a, ogg, mp4, mov — and only a regular file whose first bytes really are that format, up to 89 MB. A directory, a device, a pipe or a text file renamed to .png is refused before anything is read or sent. What does go goes to your studio (STUDIO_URL), into the account the token belongs to, and nowhere else.
HTTPS only. The token travels over https://, or over plain http:// to a studio on this machine (localhost, 127.0.0.1, [::1]). Any other http:// address is refused before a request is made.
License
MIT. This file is a copy of the studio's own mcp-server.js; the studio also serves it as longtake.studio/mcp-server.cjs.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.