mcp.film

mcp.film / Model Hubs / EmpirioLabs AI MCP

EmpirioLabs AI MCP

OfficialRemoteby EmpirioLabs.ai LLCcreditsverified Sep 15, 2026

Remote hub over 180+ hosted models: video, image, speech, 3D, plus GPUs by the hour.

What it does

EmpirioLabs' official remote MCP puts its inference platform behind one streamable-HTTP endpoint — generate_video against Kling O3, Wan 2.7, Seedance 2.0 Pro, Hunyuan Video 1.5, Pixverse V5.6 and SVI 2.0 Pro, stills from Qwen Image 2.0, Seedream 5.0 Lite, Hunyuan Image 3 and Nova Canvas, speech from GLM-TTS and Gemini TTS, music from Stable Audio 2.5, plus transcription, 3D generation and a compose_video recipe step. Useful to a filmmaker-agent as a cheap second source for open-weights video models and for the rarer surfaces most hubs skip: 3D assets and per-hour GPU instances in the same connection.

Connect

Remote endpoint (Streamable HTTP)
https://mcp.empiriolabs.ai/mcp
Claude Code
claude mcp add --transport http empiriolabs https://mcp.empiriolabs.ai/mcp
Claude Desktop (claude_desktop_config.json)
{
  "mcpServers": {
    "empiriolabs": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.empiriolabs.ai/mcp"
      ]
    }
  }
}
Cursor (.cursor/mcp.json)
{
  "mcpServers": {
    "empiriolabs": {
      "url": "https://mcp.empiriolabs.ai/mcp"
    }
  }
}
Auth: oauth — OAuth 2.1 browser sign-in for Claude and ChatGPT; scripts and editors can instead send an EmpirioLabs API key as an Authorization: Bearer header. Scopes are per-toolset (media, models, jobs, gpu, hosted_agents, …)

Sample tools

list_modelsget_modelgenerate_videogenerate_imageedit_imagecompose_videogenerate_3dtext_to_speechgenerate_audiotranscribe_audioanalyze_imageget_job

Field notes

Listed one day after launch: the repo was created 14 Sep 2026, has 0 stars and no LICENSE file, and is a README pointing at the hosted endpoint rather than server code. The endpoint itself is real — it serves OAuth protected-resource metadata naming platform.empiriolabs.ai — and the docs are unusually complete for a new service, but treat this as unproven and check pricing yourself before committing a pipeline to it. Only 13 of the 78 tools are media generation; 39 are GPU-cloud and hosted-agent plumbing that a film pipeline will never call, so scope the connection down if your client supports it. Two operational traps: generation tools block for at most 50 seconds and then hand back a job_id you have to poll via get_job, and output URLs are signed and expire, so download anything you intend to keep. Read-only catalog, job and usage tools cost nothing. No free tier — pay-as-you-go prepaid credits (crypto top-ups accepted) or weekly allowance plans from $19.90/mo, and those allowances are trivial for film work: Lite grants 20 seconds of video and 7 images per week, Max 240 seconds. Model naming is self-reported by the platform; verify a slug with list_models rather than trusting a marketing page. For anything load-bearing, fal and replicate have the track record this does not yet.

Also in Model Hubs

Pairs well with