ShortDrama Studio

Connect your AI agent

ShortDrama Studio is an MCP server. Any agent that speaks the Model Context Protocol can create a drama, write its story bible, characters, episode outlines and scene-by-scene scripts, check them, generate the images and videos, and render the episodes — in your account, with your permission.

Server URL: https://mcp.shortdrama.studio/mcp

Writing stories through your own agent never costs ShortDrama credits: your agent's tokens are yours. Credits are spent only on generation (images, video, music) and rendering.

The fastest way: one prompt

Copy this prompt and paste it into Claude Code, Codex, Cursor, OpenCode or another agent that can run commands:

Fetch and execute the appropriate instructions to set me up for ShortDrama Studio from https://shortdrama.studio/agent-setup/prompt.md

Your agent reads our setup instructions and first asks you for an access token. Create one in Settings → MCP & Agents → Access tokens → New token: give it a name, keep Read and Write, tick Generate if the agent may spend credits on pictures, video and music, pick an expiry and a daily credit cap, then copy the token (it starts with sds_pat_ and is shown once) and paste it into the chat. The agent checks it and adds the server on its own. Then restart the agent and ask it for a drama, for example "Create a 10-episode CEO romance and cast it."

Prefer to do it by hand? Create a token the same way and put it in place of <TOKEN> in the steps below.

Claude Code

claude mcp add --transport http --scope user shortdrama https://mcp.shortdrama.studio/mcp --header "Authorization: Bearer <TOKEN>"

--scope user makes the server available in every project. Then restart Claude Code.

Prefer one step? Download the Claude Code plugin: it adds the server and installs the official skills locally, and signs in with your browser (type /mcp, choose shortdrama) instead of a token. Unzip it, then in Claude Code run /plugin marketplace add ./shortdrama-marketplace and /plugin install shortdrama@shortdrama. Use the plugin or the command above, not both, or the tools show up twice.

Codex

Add to ~/.codex/config.toml:

[mcp_servers.shortdrama]
url = "https://mcp.shortdrama.studio/mcp"
http_headers = { Authorization = "Bearer <TOKEN>" }

On older Codex versions you may also need experimental_use_rmcp_client = true.

OpenCode

Add the server under "mcp" in ~/.config/opencode/opencode.json:

"shortdrama": { "type": "remote", "url": "https://mcp.shortdrama.studio/mcp", "enabled": true, "oauth": false, "headers": { "Authorization": "Bearer <TOKEN>" } }

Cursor, VS Code, Windsurf, Gemini CLI and others

Add https://mcp.shortdrama.studio/mcp as a remote MCP server (Streamable HTTP) with the header Authorization: Bearer <TOKEN>.

  • Cursor — ~/.cursor/mcp.json: { "mcpServers": { "shortdrama": { "url": "https://mcp.shortdrama.studio/mcp", "headers": { "Authorization": "Bearer <TOKEN>" } } } }
  • VS Code (GitHub Copilot) — .vscode/mcp.json (keep the token out of the project with an input): { "inputs": [{ "type": "promptString", "id": "shortdrama-token", "description": "ShortDrama Studio access token", "password": true }], "servers": { "shortdrama": { "type": "http", "url": "https://mcp.shortdrama.studio/mcp", "headers": { "Authorization": "Bearer ${input:shortdrama-token}" } } } }
  • Windsurf — ~/.codeium/windsurf/mcp_config.json: { "mcpServers": { "shortdrama": { "serverUrl": "https://mcp.shortdrama.studio/mcp", "headers": { "Authorization": "Bearer <TOKEN>" } } } }
  • Gemini CLI — ~/.gemini/settings.json: { "mcpServers": { "shortdrama": { "httpUrl": "https://mcp.shortdrama.studio/mcp", "headers": { "Authorization": "Bearer <TOKEN>" } } } }

Menu names change between versions; if a step looks different, look for "MCP server" or "connector" in your app's settings.

Claude.ai, Claude Desktop and ChatGPT

Their connectors cannot send a token, so they sign in instead: add a custom connector with the URL above (Settings → Connectors) and approve the connection in your browser when asked.

Hermes, scripts and bots

Send the token with every request:

Authorization: Bearer <TOKEN>

The same token works on the MCP server and on the REST tools at https://api.shortdrama.studio/v2/tools/<tool_name>.

About access tokens

  • Shown once when you create it; copy it then. Treat it like a password and keep it out of files you commit.
  • Each token has its own permissions, expiry (7, 30, 90 or 365 days) and daily credit cap.
  • When it expires or you revoke it, the agent gets 401: create a new one and give it to the agent.

Smaller tool sets

Some agents choose tools better when they see fewer. Two extra endpoints serve part of the tools with the same token:

  • https://mcp.shortdrama.studio/mcp/writer — reading and writing the story.
  • https://mcp.shortdrama.studio/mcp/producer — reading the story and producing it.

What your agent gets

  • The whole story in one call (get_drama_context), with overview, outline and full levels, so it never writes episode 7 without knowing episodes 1–6.
  • Every valid option (get_drama_spec, get_catalog): formats, genres, tropes, visual styles, episode lengths, shots, camera moves, languages and your own custom presets — no guessing.
  • Skills: official playbooks for the workflow, vertical-drama storycraft, genres (CEO romance, revenge, contract marriage, werewolf), visual prompting, character consistency and dialogue, plus your own. Served as MCP skills (skill://shortdrama/…) and through list_skills / get_skill for agents that don't load MCP skills yet, so there is nothing to install; a local copy is one download away (unzip into ~/.claude/skills or ~/.codex/skills). Your agent can save your house style as a skill (save_skill), import one from a public GitHub folder (import_skill), and pin skills to a drama (attach_skills) so every agent that works on it follows the same rules. Manage them on the Skills page.
  • A checker (lint_drama): hooks, cliffhangers, pacing, dialogue length and more, each finding with a fix.
  • Production: character portraits, location plates, keyframes and scene videos (Wan 3.0, MiniMax H3, HeyGen Video 1) in which the model speaks your dialogue, a fixed voice per character, episode music, and the rendered MP4 with subtitles — with estimate_cost first and wait_for_jobs instead of polling. Use the /mcp/producer endpoint for these.
  • Import and export in JSON (shortdrama.script.v1) or Fountain, for agents that like to write a whole season in a file first.
  • Live in your studio: everything your agent makes shows up at once in My Projects and the studio, where you can keep editing it yourself.

The full tool reference (every tool with its permissions and input schema) is public at api.shortdrama.studio/v2/schemas/tools.json.

You stay in control

  • You create every token and choose what it may do: read, write, use skills, spend credits.
  • Credit cap per token: the most it may spend on generation per day.
  • Everything is undoable: every change an agent makes is kept in the drama's history with the agent's name on it, and can be reverted.
  • Revoke any time in Settings → MCP & Agents; access ends at once.

FAQ

What does the setup prompt do?

It asks your agent to read our setup instructions at shortdrama.studio/agent-setup/prompt.md and follow the part for itself: ask you for an access token, check it, then add the ShortDrama Studio MCP server with one command or one config entry. It installs nothing else and spends no credits. You can open the file and read it before you paste the prompt.

Where do I get the access token?

On shortdrama.studio, in Settings → MCP & Agents → Access tokens → New token. It is shown once, so copy it before closing the panel.

Does connecting an agent cost credits?

No. Connecting and writing stories is free. Only generation (images, video, music) and rendering spend credits, within the cap you set.

Which agents work?

Any MCP client with Streamable HTTP: Claude Code, Codex, Cursor, OpenCode, VS Code, Windsurf, Gemini CLI, Claude.ai, ChatGPT and others. Apps that cannot send a token (Claude.ai, ChatGPT connectors) sign in with OAuth.

Can my agent see other people's dramas?

No. A token acts only in your own workspace, with the permissions you gave it.