Developers · Model Context Protocol

Create with Hartora
from your AI agent.

The Hartora MCP server plugs your account into Claude Code, Claude Desktop, Cursor, Codex — any client that speaks MCP. Ask your agent for an image, a video or a song; it lands in your Hartora library, billed in your credits.

npx -y @hartora/mcp   # that's the whole install

Create an API key

Go to Settings → API Keys and create a key. It is shown once — copy it right away. Keys look like hart_… and require the Starter plan or higher. Revoke a key any time from the same page.

Add Hartora to your client

Claude Code (CLI)

claude mcp add hartora -e HARTORA_API_KEY=hart_... -- npx -y @hartora/mcp

Claude Desktop — claude_desktop_config.json

{
  "mcpServers": {
    "hartora": {
      "command": "npx",
      "args": ["-y", "@hartora/mcp"],
      "env": { "HARTORA_API_KEY": "hart_..." }
    }
  }
}

Cursor — .cursor/mcp.json (same shape as Claude Desktop)

Codex CLI — ~/.codex/config.toml

[mcp_servers.hartora]
command = "npx"
args = ["-y", "@hartora/mcp"]
env = { "HARTORA_API_KEY" = "hart_..." }

Requires Node.js 18+. Restart the client after editing its config.

Ask for something

"Generate an image of a deer leaping over ocean waves, minimal logo style"
"Make a 5-second video of rain on a neon street"
"Write and produce a lo-fi song about summer"

Images usually return in under a minute, inline. Video and music return a job_id immediately — your agent checks back with get_job after the suggested delay (videos take 1–5 minutes). Everything also appears in your Hartora library.

Six tools

generate_image	Text → image(s). Waits up to ~45s and returns image URLs directly; batches return a job_id.
generate_video	Text/image → video. Returns a job_id immediately — videos take 1–5 minutes.
generate_music	Prompt or lyrics → full songs via Suno. Returns a job_id.
get_job	Check a job by id — status, progress, result URLs when finished.
list_models	List available image/video/music models with their credit costs.
get_credits	Your current credit balance and plan.

Billed in credits

Each generation costs the same credits as in the app — see list_models or the pricing page. Failed jobs are auto-refunded.

Safe to retry

Job creation is idempotent: a network retry can never charge you twice for one request.

Rate limits

Max 10 job submissions per minute, and concurrent jobs follow your plan. The server tells your agent when to back off.

Ready to build?

Create your API key and your agent is 60 seconds away from its first image.

Get your API key