@simpligen/mcp

MCP server that lets an AI agent drive SimpliGen (local + cloud AI image/video generation). Requires the SimpliGen desktop app running and a pairing token (Settings -> Connect an agent).

Configure (Claude Code / Claude Desktop)

The easiest path is in the SimpliGen app: Settings -> Connect an agent, pick your client, and click Install (or copy the guided command). The manual config below is the fallback.

{
  "mcpServers": {
    "simpligen": {
      "command": "npx",
      "args": ["-y", "@simpligen/mcp"],
      "env": { "SIMPLIGEN_TOKEN": "sg-agent-..." }
    }
  }
}

Tools

list_capabilities, get_status, list_projects, create_project, upload_file, generate, get_job, wait_for_result, list_jobs, cancel_job, prepare_preset, get_result_image.

get_result_image returns a completed job's result image inline (a viewable image block) so the agent can display it in chat. Claude clients cap a tool result at ~1MB and only render base64 image blocks, so large renders are re-encoded to a full-resolution WebP preview (same dimensions; quality steps down, and only very large images are resized) that fits the cap. The uncompressed original file path is always included in the caption, so the agent keeps the full-quality artifact. Video results return the file path instead.

Env

• SIMPLIGEN_TOKEN (required) -- pairing token from the app.
• SIMPLIGEN_API_PORT (optional) -- override the control-API port (else read from the app's discovery file).
• SIMPLIGEN_USERDATA_DIR / SIMPLIGEN_APP_NAME (optional) -- locate the app's data dir (CN variant / custom).

Manual smoke (for maintainers)

With the app running and a token issued:

SIMPLIGEN_TOKEN=<token> npx @modelcontextprotocol/inspector node src/bin.js

Verify: tool list appears; call get_status; call list_capabilities; call upload_file with a PNG path and confirm a handle is returned; call generate with a local preset and confirm a jobId; call wait_for_result and confirm a resultPath; call get_result_image and confirm the image renders inline.