MCP Web UI is a web-based user interface that serves as a Host within the Model Context Protocol (MCP) architecture. It provides a powerful and user-friendly interface for interacting with Large Language Models (LLMs) while managing context aggregation and coordination between clients and servers.

MCP Web UI is designed to simplify and enhance interactions with AI language models by providing:

• A unified interface for multiple LLM providers
• Real-time, streaming chat experiences
• Flexible configuration and model management
• Robust context handling using the MCP protocol

• 🤖 Multi-Provider LLM Integration:
  • Anthropic (Claude models)
  • OpenAI (GPT models)
  • Ollama (local models)
  • OpenRouter (multiple providers)
• 💬 Intuitive Chat Interface
• 🔄 Real-time Response Streaming via Server-Sent Events (SSE)
• 🔧 Dynamic Configuration Management
• 📊 Advanced Context Aggregation
• 💾 Persistent Chat History using BoltDB
• 🎯 Flexible Model Selection

• Go 1.23+
• Docker (optional)
• API keys for desired LLM providers

1. Clone the repository:

   git clone https://github.com/MegaGrindStone/mcp-web-ui.git
   cd mcp-web-ui

2. Configure your environment:

   mkdir -p $HOME/.config/mcpwebui
   cp config.example.yaml $HOME/.config/mcpwebui/config.yaml

3. Set up API keys:

   export ANTHROPIC_API_KEY=your_anthropic_key
   export OPENAI_API_KEY=your_openai_key
   export OPENROUTER_API_KEY=your_openrouter_key

go mod download
go run ./cmd/server/main.go

docker build -t mcp-web-ui .
docker run -p 8080:8080 \
  -v $HOME/.config/mcpwebui/config.yaml:/app/config.yaml \
  -e ANTHROPIC_API_KEY \
  -e OPENAI_API_KEY \
  -e OPENROUTER_API_KEY \
  mcp-web-ui

The configuration file (config.yaml) provides comprehensive settings for customizing the MCP Web UI. Here's a detailed breakdown:

• port: The port on which the server will run (default: 8080)
• logLevel: Logging verbosity (options: debug, info, warn, error; default: info)
• logMode: Log output format (options: json, text; default: text)

• systemPrompt: Default system prompt for the AI assistant
• titleGeneratorPrompt: Prompt used to generate chat titles

The llm section supports multiple providers with provider-specific configurations:

• provider: Choose from: ollama, anthropic, openai, openrouter
• model: Specific model name (e.g., 'claude-3-5-sonnet-20241022')
• parameters: Fine-tune model behavior:
  • temperature: Randomness of responses (0.0-1.0)
  • topP: Nucleus sampling threshold
  • topK: Number of highest probability tokens to keep
  • frequencyPenalty: Reduce repetition of token sequences
  • presencePenalty: Encourage discussing new topics
  • maxTokens: Maximum response length
  • stop: Sequences to stop generation
  • And more provider-specific parameters

• Ollama:

  • host: Ollama server URL (default: http://localhost:11434)

• Anthropic:

  • apiKey: Anthropic API key (can use ANTHROPIC_API_KEY env variable)
  • maxTokens: Maximum token limit
  • Note: Stop sequences containing only whitespace are ignored, and whitespace is trimmed from valid sequences as Anthropic doesn't support whitespace in stop sequences

• OpenAI:

  • apiKey: OpenAI API key (can use OPENAI_API_KEY env variable)
  • endpoint: OpenAI API endpoint (default: https://api.openai.com/v1)
  • For using alternative OpenAI-compatible APIs, see this discussion thread

• OpenRouter:

  • apiKey: OpenRouter API key (can use OPENROUTER_API_KEY env variable)

The genTitleLLM section allows separate configuration for title generation, defaulting to the main LLM if not specified.

• mcpSSEServers: Configure Server-Sent Events (SSE) servers

  • url: SSE server URL
  • maxPayloadSize: Maximum payload size

• mcpStdIOServers: Configure Standard Input/Output servers

  • command: Command to run server
  • args: Arguments for the server command

SSE Server Example:

mcpSSEServers:
  filesystem:
    url: https://yoursseserver.com
    maxPayloadSize: 1048576 # 1MB

StdIO Server Examples:

1. Using the official filesystem MCP server:

mcpStdIOServers:
  filesystem:
    command: npx
    args:
      - -y
      - "@modelcontextprotocol/server-filesystem"
      - "/path/to/your/files"

This example can be used directly as the official filesystem mcp server is an executable package that can be run with npx. Just update the path to point to your desired directory.

2. Using go-mcp filesystem MCP server:

mcpStdIOServers:
  filesystem:
    command: go
    args:
      - run
      - github.com/your_username/your_app # Replace with your app
      - -path
      - "/data/mcp/filesystem" # Path to expose to MCP clients

For this example, you'll need to create a new Go application that imports the github.com/MegaGrindStone/go-mcp/servers/filesystem package. The flag naming (like -path in this example) is completely customizable based on how you structure your own application - it doesn't have to be called "path". This example is merely a starting point showing one possible implementation where a flag is used to specify which directory to expose. You're free to design your own application structure and command-line interface according to your specific needs.

port: 8080
logLevel: info
systemPrompt: You are a helpful assistant.

llm:
  provider: anthropic
  model: claude-3-5-sonnet-20241022
  maxTokens: 1000
  parameters:
    temperature: 0.7

genTitleLLM:
  provider: openai
  model: gpt-3.5-turbo

• cmd/: Application entry point
• internal/handlers/: Web request handlers
• internal/models/: Data models
• internal/services/: LLM provider integrations
• static/: Static assets (CSS)
• templates/: HTML templates

1. Fork the repository
2. Create a feature branch
3. Commit your changes
4. Push and create a Pull Request

MIT License