MCP Inception

🚀 MCP-Inception Server (MCP Server)

Call another MCP client from your MCP client. Assign tasks and offload context windows. It's an agent within an agent!

This is a TypeScript-based MCP server that implements a simple LLM query system.

• Act as both an MCP server and client in one program.
• Built with mcp-client-cli.
• Offload context windows.
• Assign tasks.
• Execute tasks in parallel and perform map-reduce operations.

✨ Features

Tools

• execute_mcp_client - Pose a question to another LLM, ignore all intermediate steps of its tool execution, and return the output.
  • Accept the question as a required parameter.
  • Return the answer, ignoring all intermediate context.
• execute_parallel_mcp_client - Accept a list of inputs and a main prompt, and execute the prompt in parallel for each string in the input. For example, get the current time in 6 major cities - London, Paris, Tokyo, Rio, New York, Sydney.
  • Use "What's the time in this city?" as the main prompt.
  • Use London, Paris, etc. as the input list.
  • Run the prompt for each input in parallel.
  • Note: Please wait until this is resolved before using this feature.
• execute_map_reduce_mcp_client - Process multiple items in parallel and then sequentially reduce the results to a single output.
  • Use mapPrompt with the {item} placeholder for single-item processing.
  • Use reducePrompt with the {accumulator} and {result} placeholders for result combination.
  • Process the list of items to be analyzed.
  • Optionally provide an initialValue for the accumulator.
  • Process items in parallel and then sequentially reduce the results.
  • Example use case: Analyze multiple documents and then synthesize a summary of key insights from all documents.

🚀 Quick Start

Development

Dependencies:

• Install mcp-client-cli.
  • Install the configuration file simultaneously and set the required MCP server in ~/.llm/config.json.
• Create a bash script file, activate the virtual environment, and execute the llm executable.

#!/bin/bash
source ./venv/bin/activate
llm --no-confirmations

Install Packages

Install dependencies:

npm install

Build the server:

npm run build

Develop with automatic rebuild:

npm run watch

📦 Installation

To use it with Claude Desktop, add the server to the configuration file:

On MacOS: ~/Library/Application Support/Claude/claude_desktop_config.json On Windows: %APPDATA%/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "mcp-inception": {
      "command": "node",
      "args": ["~/Documents/Cline/MCP/mcp-inception/build/index.js"], // build/index.js of this repository
      "disabled": false,
      "autoApprove": [],
      "env": {
        "MCP_INCEPTION_EXECUTABLE": "./run_llm.sh", // The bash file in Development -> Dependencies
        "MCP_INCEPTION_WORKING_DIR": "/mcp-client-cli working directory"
      }
    }
  }
}

Debugging

Since the MCP server communicates via stdio, debugging can be challenging. We recommend using MCP Inspector, which is available as a program script:

npm run inspector

The Inspector will provide a URL to access the browser debugging tools.