Runware

You can run the Runware MCP Server to access fast image and video generation tools through a structured MCP experience. It supports local MCP installation for Claude Desktop and containerized SSE-based operation, giving you real-time interactions and seamless model orchestration with Runware’s AI services.

How to use

To use the Runware MCP Server, install the local MCP client and run the provided stdio command to start the server via Claude Desktop. You’ll then connect your MCP client to the server, sending requests to perform image or video generation tasks, model discovery, and utility operations like uploading images. The server validates inputs to prevent problematic uploads and provides clear errors when something goes wrong. High-quality results come from choosing appropriate models for image and video generation and leveraging the automatic model selection and validation features.

How to install

Prerequisites you need before installation include Python 3.10 or higher and an active Runware API key. You’ll also want a clean environment for Python dependencies.

# 1) Ensure you have Python installed (>= 3.10).
# 2) Create and activate a virtual environment (uv is recommended):
uv venv
source .venv/bin/activate

# 3) Install the MCP package from the local project directory (as shown):
mcp install --with-editable . runware_mcp_server.py

Configuration and environment

Set up your environment and API access to Runware before starting the server.

RUNWARE_API_KEY=your_api_key_here

Starting the server via Claude Desktop (stdio)”

Use the following command to install and start the Runware MCP Server locally from Claude Desktop. This runs the MCP server as a local process and exposes the configured MCP endpoint for your client.

mcp install --with-editable . runware_mcp_server.py

Deployment options

You can run the server in a container for SSE-based production use or install it directly for MCP-based desktop workflows.

# Docker SSE deployment (example, if you are containerizing this server):
docker build -t runware_mcp_sse .
docker run --rm -p 8081:8081 runware_mcp_sse

Model and tool usage

The server provides a suite of image and video generation tools, as well as utility capabilities to manage models and uploads. Choose the right tool for your task and provide appropriate inputs to get fast, accurate results.

Security and notes

• Ensure your API key remains secret and is not committed to public repositories.
• The server performs input validation to prevent problematic data from affecting performance or context windows.
• Use HTTPS or other secure transport for production deployments and safeguard your API endpoints.

Troubleshooting

If you encounter errors, check that RUNWARE_API_KEY is correctly set, that the MCP server process is running, and that your client is configured to connect to the correct endpoint. Review any error messages for hints about input validation or model availability.

Notes

All tools and endpoints are designed to work with the Runware MCP Server. You can discover available image and video models, perform image uploads to obtain Runware UUIDs, and search for models directly through the MCP client once the server is running.

Available tools

imageInference

Full-featured image generation with advanced parameters to create high-quality visuals.

photoMaker

Subject personalization using PhotoMaker technology to tailor outputs.

imageUpscale

Increase image resolution with quality-preserving upscaling.

imageBackgroundRemoval

Remove backgrounds using multiple AI models for clean composites.

imageCaption

Generate descriptive captions for images using AI.

imageMasking

Automatic mask generation for faces, hands, and people.

videoInference

Text-to-video and image-to-video generation capabilities.

listVideoModels

Discover available video models for your projects.

getVideoModelInfo

Retrieve detailed specifications for a selected video model.

imageUpload

Upload local images to obtain Runware UUIDs for processing.

modelSearch

Search and discover AI models available on the platform.