MCP server
The MCP server lets an AI assistant work in your Evocal account: send a video, start the dub, check and fix segments, and download the files. It uses your API key, with the same credits and rules as the app.
Before you start
- The MCP server is part of the Pro, Studio and Enterprise plans, like the API.
- Make a key in the app, under Account, API. Give the assistant a key of its own, so you can revoke it without stopping anything else.
- Your assistant must let you add a remote MCP server with a header. Agents that work in a terminal and code editors do; chat apps that connect only by signing in cannot use it yet.
Connecting
The server's address is https://www.evocal.app/api/mcp (Streamable HTTP). It takes the key in the Authorization header. For example, in Claude Code:
claude mcp add --transport http evocal https://www.evocal.app/api/mcp \
--header "Authorization: Bearer evk_..."Editors that keep their servers in a JSON file take an entry like this one. Check your editor's guide for the file's name and its exact form.
{
"mcpServers": {
"evocal": {
"url": "https://www.evocal.app/api/mcp",
"headers": { "Authorization": "Bearer evk_..." }
}
}
}Then ask, for example: “Dub lesson-4.mp4 from Portuguese into English, and tell me what it costs before you start.”
What the assistant can do
| Tool | What it does |
|---|---|
get_account | The plan and the credits now. |
list_languages | The language pairs open now. |
list_projects | Your projects, newest first. |
get_project | One project, with its file's state and its dubs. |
create_project | Creates a project with one dub for each target language. |
start_upload | Opens the file's upload and gives the first parts' addresses. |
get_upload_part_urls | Addresses for more parts, or new ones for parts that failed. |
complete_upload | Ends the upload; the file is then checked. |
start_dub | Starts a dub. Spends credits. |
get_job | A job's state and progress. |
get_dub | One dub, with the last job that ran on it. |
get_segments | The dub's segments: times, texts and notices. It can show only the ones with notices or edits. |
edit_segment | Saves a new text for a segment. |
quote_regeneration | What regenerating some segments costs. |
regenerate_segments | Regenerates segments with their saved texts. Spends credits. |
finish_dub | Finishes the editing and opens the downloads. |
get_download_links | Links to the dub's files, good for 5 minutes. |
get_subtitles | The subtitles as text, in SRT or WebVTT. |
delete_project | Deletes a project with its files and dubs. |
Sending the file
The assistant reads the file on your computer and sends it in parts, so it must run where the file is, with a terminal or access to your files. It cuts the file into parts of part_size bytes and sends each one with a PUT to its address.
An assistant that cannot reach your files can still work on the projects you send in the app: read the segments, fix them, regenerate and download.
Credits and safety
- The assistant spends your credits: a dub costs one credit per second of video (at least 30), a regeneration the seconds of its segments (at least 15). The server tells it to give you the cost and wait for your yes before either, but you answer for every call made with your key.
- Finishing a dub and deleting a project cannot be undone. The assistant is told to ask you first.
- Every file follows the same rules as in the app: send only voices that are your own or whose speakers gave you written permission. The key's voice confirmation covers each file.
- Its calls count in the API's limits: 120 a minute and 3,600 an hour for the whole account.
- The assistant sees what the API shows: your projects, dubs, segments and credits. If the key leaks, revoke it in Account, API.