Twosplit MCP Server
# Twosplit MCP Server
An MCP server that leverages multiple Claude instances to provide enhanced responses. It sends the same prompt to two separate instances of Claude and uses a third instance to combine or select the best elements from both responses.
## Features
- Supports multiple Claude models:
- claude-3-opus-latest
- claude-3-5-sonnet-latest
- claude-3-5-haiku-latest
- claude-3-haiku-20240307
- Gets single, direct responses from each AI
- Shows original responses and source attribution
- Returns optimized final response
## Installation
1. Clone the repository
2. Install dependencies:
```bash
npm install
```
3. Build the server:
```bash
npm run build
```
## Configuration
The server requires an Anthropic API key to function. Set it as an environment variable:
```bash
export ANTHROPIC_API_KEY=your-api-key-here
```
## Usage
The server provides a single tool called `twosplit` with the following parameters:
- `prompt` (required): The prompt to send to Claude
- `model` (required): The Claude model to use (must be one of the supported models listed above)
Example tool usage in Claude:
```
<use_mcp_tool>
<server_name>twosplit</server_name>
<tool_name>twosplit</tool_name>
<arguments>
{
"prompt": "Write a short story about a robot learning to paint",
"model": "claude-3-5-sonnet-latest"
}
</arguments>
</use_mcp_tool>
```
The response will include:
1. The final optimized response
2. Original responses from both AIs
3. Source attribution showing which parts came from which AI
## How it Works
1. The server sends the same prompt to two separate instances of the specified Claude model, requesting a single direct response
2. A third instance analyzes both responses and either:
- Selects the single best response if one is clearly superior
- Creates a new response that combines the best elements from both responses
3. The final response, original responses, and source attribution are all included in the output
## Development
To run the server in watch mode during development:
```bash
npm run watch
```
To inspect the server's capabilities:
```bash
npm run inspector
TDQS
Scored across 1 tool
With only one tool, there is no possibility of ambiguity or overlap between tools. The tool's purpose is clearly defined as getting multiple AI perspectives and combining them, which is distinct and unambiguous in this single-tool context.
Since there is only one tool, naming consistency is inherently perfect. The tool name 'twosplit' follows a single pattern with no deviations or mixing of conventions to evaluate.
A single tool is generally too few for most server purposes, as it limits functionality and flexibility. For a server named 'Twosplit MCP Server' that aims to provide multiple AI perspectives, one tool feels thin and under-scoped, potentially leaving gaps in operations like configuration or result refinement.
The tool surface is severely incomplete for the implied domain of handling AI perspectives. While the 'twosplit' tool covers the core action, there are obvious gaps such as no tools for managing perspectives, adjusting combination parameters, or retrieving historical results, which could lead to agent failures in more complex workflows.