SwarmUI_MCP_Agent

What is SwarmUI_MCP_Agent?

SwarmUI_MCP_Agent is a prototype MCP (Model Context Protocol) server that integrates with SwarmUI's local API and a LangChain Ollama agent. It exposes SwarmUI's /GenerateText2Image and /GetNewSession endpoints as tools, allowing an LLM agent to generate images from text prompts. The server is built with Node.js, TypeScript, Express, and Axios, while the agent uses Python, LangChain, and Ollama.

How to use SwarmUI_MCP_Agent?

From the SwarmUI_MCP_Agent/src directory, run tsc server.ts && node server.js to start the MCP server. Then ensure SwarmUI is running, and from the root directory run python run.py to launch the Streamlit UI. In the UI, enter the SwarmUI port, select an LLM model, type an image prompt, and click submit.

Key features of SwarmUI_MCP_Agent

• Exposes SwarmUI image generation as MCP tools
• Integrates a local Ollama agent (dolphin-mistral)
• Provides a Streamlit UI for prompt entry
• Automatically fetches a new session and generates images
• Generated images saved in SwarmUI's Output directory

Use cases of SwarmUI_MCP_Agent

• Generating images from natural language prompts via a local agent
• Experimenting with MCP and LangChain tool‑calling workflows
• Prototyping an AI‑assisted image generation pipeline
• Offline image generation using a local SwarmUI instance
• Testing MCP server integration with Ollama

FAQ from SwarmUI_MCP_Agent

What does SwarmUI_MCP_Agent do?

It acts as a bridge between an LLM agent (running on Ollama) and the SwarmUI image generation API. The agent receives a text prompt, calls the SwarmUI /GenerateText2Image endpoint, and returns the generated image.

What are the runtime dependencies?

You need Node.js, npm, Python, pip, and an Ollama installation with a model (e.g., dolphin‑mistral). SwarmUI must be running locally.

How does the server communicate with SwarmUI?

The MCP server creates an Axios HTTP client that sends POST requests to the SwarmUI API on localhost (default port configurable in code). No authentication is mentioned.

Are there any known limitations?

The prototype only exposes the /GenerateText2Image endpoint. The model and port are hardcoded and must be changed manually. The agent has a max_iterations of 1, and Pydantic input validation is not yet fully functional for multiple inputs.

Where are generated images stored?

Images are saved in SwarmUI's /Output directory and can also be viewed in SwarmUI's Image History.