# Runware MCP [Health: Active]

**Category:** 🎥 Multimedia Process  
**Repository:** https://github.com/Runware/mcp  
**GitHub Stars:** 0  
**npm Downloads (last month):** 1473  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/mcp-203

## Description
Access the full Runware API as MCP tools: image, video, audio, 3D, and text.

## Tools
Capabilities this server exposes over MCP:

- **run** — execute any Runware inference task (image, video, audio, 3D, upscaling, captioning, etc.) on a given model
- **model_schema** — fetch the parameter schema for a specific model
- **list_models** — list Runware's official, curated model integrations (supports `capability`, `category`, `creator`, `search` filters)
- **model_details** — get full metadata for a curated model by AIR
- **model_examples** — get sample input/output pairs for a curated model
- **model_pricing** — get pricing overview + per-configuration examples for a curated model
- **list_capabilities** — list every model capability (e.g. `io:text-to-image`, `op:upscale`) with labels
- **model_search** — search the community model catalog (Civitai fine-tunes, custom uploads)
- **media_storage** — store media (images, video, audio, 3D models) for use as input, or delete stored media
- **model_upload** — upload a custom model
- **account** — retrieve account information including balance and usage
- **get_task_details** — retrieve the original request and response for a previous task

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "runware-mcp": {
    "command": "npx",
    "args": ["-y","@runware/mcp"],
    "env": {
      "RUNWARE_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `RUNWARE_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What Runware MCP MCP server does

Runware MCP MCP server connects an MCP-compatible agent to the Runware API. It covers inference tasks across image, video, audio, 3D, upscaling, background removal, captioning, and other capabilities exposed by Runware models. The agent can ask for a result in natural language, while the server provides the API operations needed to select a model and submit the request.

The server also exposes discovery and account functions. An agent can inspect available capabilities, browse Runware’s curated model integrations, search the community catalog, review pricing, check account usage, and retrieve details from an earlier task.

## How it works

The MCP process communicates with the agent over stdio. After connection, the agent chooses among the available tools based on the request. For example, it can use `list_models` to filter official integrations by capability, category, creator, or search term, then use `model_schema` to learn the parameters accepted by a selected model.

Before submission, parameters are checked against the model’s schema. This gives the agent feedback when a value does not match the selected model. Inference is submitted through `run`, which supports tasks such as generation, upscaling, and captioning on a specified model. Results from previous work can be inspected with `get_task_details`.

## Setup and configuration

Install the package with either of these approaches:

- Run it directly with `npx -y @runware/mcp`.
- Install it globally with `npm install -g @runware/mcp`, then use the `runware-mcp` command.

A Runware API key from runware.ai is required. Set it in the `RUNWARE_API_KEY` environment variable. Claude Desktop can launch the server with the npx command and this environment variable; Cursor adds the same command and environment configuration through Settings → MCP → Add server. Codex can use the same MCP configuration.

## Tools and capabilities

Runware MCP MCP server provides these tool groups:

- `run` submits an inference task for a selected model.
- `model_schema` returns a model’s parameter schema.
- `list_models`, `model_details`, `model_examples`, and `model_pricing` describe curated Runware integrations.
- `list_capabilities` lists capability identifiers and labels.
- `model_search` searches community models, including Civitai fine-tunes and custom uploads.
- `media_storage` stores or deletes images, video, audio, and 3D media for use as input.
- `model_upload` uploads a custom model.
- `account` returns account balance and usage information.
- `get_task_details` returns the original request and response for an earlier task.

The model tools help an agent compare available integrations before calling `run`. Media storage is available when an input asset needs to be retained for a task.

## Limitations and notes

The server requires a valid Runware API key, and the available operations depend on the Runware API and its models. Model parameters are not uniform, so the agent should inspect the relevant schema before submitting a task. The package is documented for npm-based installation and stdio use; the provided setup examples cover Claude Desktop, Cursor, and Codex.

_Full upstream README: https://allmcps.com/mcp/mcp-203/readme_