Documentation
AI agents and MCP
Connect Claude, Cursor or any agent to Pixamake.
Pixamake is designed to be used by AI agents as well as people:
- an MCP server at
https://pixamake.ai/mcp(Streamable HTTP); /llms.txtand/llms-full.txt: this documentation in plain Markdown;- an OpenAPI 3.1 spec for tool generation.
Connect the MCP server
Create an API key in API and webhooks, then add the server to your client.
Claude Code
claude mcp add --transport http pixamake https://pixamake.ai/mcp \
--header "Authorization: Bearer YOUR_API_KEY"Cursor (~/.cursor/mcp.json)
{
"mcpServers": {
"pixamake": {
"url": "https://pixamake.ai/mcp",
"headers": { "Authorization": "Bearer YOUR_API_KEY" }
}
}
}Claude Desktop and other stdio clients
{
"mcpServers": {
"pixamake": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://pixamake.ai/mcp", "--header", "Authorization: Bearer YOUR_API_KEY"]
}
}
}Tools
| Tool | What it does |
|---|---|
get_account |
Plan, credit balance and limits |
estimate_credits |
Credits for a video duration |
create_upload |
Presigned URL to upload a local file (the agent then sends the bytes with HTTP PUT) |
create_job |
Start cleaning a video from video_url or upload_id |
get_job |
Status, progress and download link |
wait_for_job |
Waits up to 50 seconds for a job to finish, then returns it |
list_jobs |
Recent jobs of the workspace |
cancel_job |
Stop a job and get the credits back |
Example request to an agent: "Clean the captions from https://cdn.example.com/ad.mp4 and give me the download link."
Guidelines for agents
- Check the balance with
get_accountbefore starting large batches, and tell the user when credits run out (402 insufficient_credits). - Reuse the same
Idempotency-Keywhen retrying a creation over HTTP. - Poll a running job every 5 to 10 seconds, or call
wait_for_jobrepeatedly. Most 30 second videos take a few minutes. result.urlexpires after one hour: download it, or fetch the job again for a new link.- Never print the API key in answers or logs.