RunAPI MCP Server

What it does

RunAPI's official MCP server gives agents catalog discovery, input-schema inspection, prompt search, pricing lookup, media task creation, polling and balance checks across its model marketplace. It now ships in two forms: a hosted remote at mcp.runapi.ai and the local stdio npm package. For filmmakers it is a pragmatic aggregator for Flux, Kling, Seedance, Veo, Suno, ElevenLabs and related media endpoints when you want the agent to pick a model from current constraints instead of memorized slugs.

Connect

https://mcp.runapi.ai/mcp

claude mcp add runapi -s user -- npx -y @runapi.ai/mcp

{
  "mcpServers": {
    "runapi": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.runapi.ai/mcp"
      ]
    }
  }
}

{
  "mcpServers": {
    "runapi": {
      "url": "https://mcp.runapi.ai/mcp"
    }
  }
}

Sample tools

Field notes

Apache-2.0, npm @runapi.ai/mcp 0.15.0 published Sep 30 2026, Node >=22, 55 stars — still shipping several releases a month, with the eight-plus-login tool list unchanged. The behaviour change that matters: since 0.15.0 the server no longer validates arguments locally (model, enum, range, required-field or cross-field checks), so a bad call now reaches the service and comes back as its error rather than being caught before spend — tool descriptions still list the known values, so have the agent read get_model_info first. The hosted remote at https://mcp.runapi.ai/mcp is live (POST-only; a plain GET returns 405) and takes OAuth or a bearer key; 0.13.0 also serves the same authenticated transport from https://runapi.ai/.well-known/mcp, but the README still documents the mcp.runapi.ai URL. Hosted exposes the eight business tools; `login` stays local-only. Discovery tools run without a key because the catalog is embedded at build time, so slugs go stale between releases — pricing is fetched live, the catalog is not, and the README tells agents to call check_pricing rather than trust its own examples. create_task requires a caller-supplied idempotency key (1-512 chars), one per logical task; the hosted wait holds an SSE stream for up to 300s, after which you poll get_task instead of resubmitting. `npx @runapi.ai/mcp init claude|cursor|vscode|windsurf|roo` writes client config if you would rather not hand-edit.

Also in Model Hubs

G GenMagic MCPGenMagic Official

One hosted endpoint over 450 models: text-to-video from Veo, Kling, Seedance and Hailuo on one pay-as-you-go balance.

text-to-video · text-to-imageremote · credits

fal MCP Serverfal.ai Official

One hosted endpoint to search, run and chain 1,000+ image, video and audio models.

text-to-video · image-to-videoremote · credits

EmpirioLabs AI MCPEmpirioLabs.ai LLC Official

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

text-to-video · image-to-videoremote · credits

inference.sh MCP Serverinference.sh Official

Official hosted hub: run any image, video, audio or 3D app, then poll the task.

text-to-image · text-to-videoremote · credits

Kie.ai MCP Serverfelores

Run Veo, Seedance, Kling, Runway, Suno, ElevenLabs, FLUX and more through one local MCP.

model-hub · text-to-videocredits

Runware MCPRunware Official

Official Runware MCP for image, video, audio, 3D, upscaling, and model search.

text-to-image · text-to-videocredits

Used in

AI Commercial Sprint · Model Shootout · Make a polished commercial fast · Compare video and image models before spending the budget

Pairs well with

AKOOL MCPAKOOL Official

Hosted MCP for AKOOL face swap, character swap, image-to-video, TTS and avatar lookup.

face-swap · character-swapremote · credits

Apatero Studio MCPApatero Official

Studio MCP for talking avatars, locked recurring characters, and batched output.

talking-head · avatar-videoremote · credits

bitHuman MCPbitHuman Official

Official bitHuman CLI server: build avatar agents from a face image and render WAV-driven talking video locally.

avatar-video · lip-syncremote · credits