MCP server

Acteme runs a Model Context Protocol server, so AI agents such as Claude Code, Claude Desktop, Cursor and Codex can work in your projects directly. An agent can import a script, cast characters, fix tags flagged by the linter, estimate and render lines, and pick takes, under the same rules and with the same credits as the studio.

Connect

The server speaks Streamable HTTP at:

https://acteme.app/mcp

It authenticates with an API key sent as a bearer token. The Developer page in the studio has ready-made snippets for each client, with your key filled in.

claude mcp add --transport http acteme https://acteme.app/mcp \
  --header "Authorization: Bearer ak_live_…"

Run it in your project folder, or add --scope user to use it in every project.

mcp.json

{
  "mcpServers": {
    "acteme": {
      "url": "https://acteme.app/mcp",
      "headers": { "Authorization": "Bearer ak_live_…" }
    }
  }
}

.vscode/mcp.json

{
  "servers": {
    "acteme": {
      "type": "http",
      "url": "https://acteme.app/mcp",
      "headers": { "Authorization": "Bearer ak_live_…" }
    }
  }
}

export ACTEME_API_KEY=ak_live_…
codex mcp add acteme --url https://acteme.app/mcp \
  --bearer-token-env-var ACTEME_API_KEY

Claude Desktop runs local (stdio) servers, so it connects through mcp-remote, which needs Node.js. Open Settings → Developer → Edit Config:

claude_desktop_config.json

{
  "mcpServers": {
    "acteme": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://acteme.app/mcp", "--header", "Authorization:${ACTEME_AUTH}"],
      "env": { "ACTEME_AUTH": "Bearer ak_live_…" }
    }
  }
}

Windsurf, Gemini CLI and any other client that supports Streamable HTTP work the same way: point it at the URL and send the key in the Authorization header.

Scopes decide the tools

An agent only sees the tools its key's scopes allow, so it never tries something that would be rejected.

read alone: browse projects, lines, jobs and takes, and get the manifest.
read + write: also edit scripts, casting and the lexicon, and select takes. It can't spend credits.
Add generate to let it render, translate and estimate costs.

For an agent you're still getting to know, start without generate.

Tools

Account & voices	What it does	Scopes
`get_account`	Current user, plan limits and credit balances.	read
`list_voices`	Studio voices and Acteme Specials.	read
`search_voice_library`	Search the voice library by language, gender, pitch or free text.	read

Projects & cast	What it does	Scopes
`list_projects`	Projects with line, rendered and character counts.	read
`get_project`	Settings, characters, scenes, status counts and recent jobs.	read
`create_project`	Create a project (defaults: en-US, Pro, WAV 48 kHz).	write
`update_project`	Change locales, model, file template or audio settings.	read + write
`create_character`	Add a character and cast a voice.	write
`update_character`	Recast or restyle a character; affected lines go stale.	write
`delete_character`	Delete a character; its lines become unassigned.	write
`create_scene`	Add a scene to group lines.	write

Script	What it does	Scopes
`list_lines`	Lines in script order with status, takes and lint findings; pages with nextOffset.	read
`add_line`	Create one line.	write
`update_line`	Edit text, style or character, or approve a line.	write
`bulk_edit_lines`	Delete, assign character or scene, set style or approve many lines.	write
`move_line`	Move a line before or after another line.	write
`import_script`	Import CSV, JSON or Yarn content, with an optional dry run.	write
`set_translation`	Write a human translation for one line.	write
`translate_lines`	Machine-translate lines into target locales (spends credits).	generate
`get_lexicon`	Pronunciation substitutions.	read
`set_lexicon_entry`	Add or update a respelling; affected lines go stale.	write
`delete_lexicon_entry`	Remove a respelling.	write

Rendering	What it does	Scopes
`estimate_render`	Item count and credit cost, without queueing.	generate
`render_lines`	Queue a batch render and reserve credits.	generate
`list_jobs`	Recent render jobs, newest first.	read
`get_job`	Job status, item counts and failures.	read
`wait_for_job`	Long-poll a job until it finishes or the wait budget runs out.	read
`cancel_job`	Cancel queued items and refund their credits.	generate

Takes & delivery	What it does	Scopes
`list_takes`	All takes of a line; marks the selected one.	read
`select_take`	Make a take the one that ships.	write
`get_take_audio_url`	Short-lived signed URL for a take's audio.	read
`get_manifest`	Selected takes with file paths, sha256 and signed URLs.	read

One-off speech	What it does	Scopes
`generate_speech`	Render one line outside a project.	generate + read
`generate_dialogue`	Render a two-speaker exchange as one file.	generate + read
`get_speech_result`	Keep waiting for a pending one-off render.	read

Tools follow the same rules, credits and limits as the studio, with results kept compact for agents. Errors come back as tool errors with a stable error code, so the agent can react to insufficient_credits or a validation failure.

wait_for_job and the one-off speech tools wait up to 25 seconds per call, since most clients time tools out after about a minute. If the work isn't done yet, they say so and the agent calls again.

Credits

render_lines, translate_lines, generate_speech and generate_dialogue spend credits. Agents are asked to call estimate_render and report the cost before large renders, and most clients also ask you to approve each tool call.

Things to ask

"Import dialogue.csv into Ashfall, cast Varga with a low, gravelly voice and render her lines."
"Find lines with stage directions in brackets and rewrite them with vocal tags and styles."
"Render everything stale in German, then tell me which lines failed."
"Add a lexicon entry so 'Xyl' is read as 'Zile', and re-render the affected lines."