Disclaimer

Ok this is a difficult one. Will take some setting up unfortunately. However, if you manage to make this more straightforward, please send me PR’s.

mcp-inception MCP Server

Call another mcp client from your mcp client. Delegate tasks, offload context windows. An agent for your agent!

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

• MCP Server and Client in one
• Made with use of mcp-client-cli
• Offload context windows
• Delegate tasks
• Parallel and map-reduce execution of tasks

Features

Tools

• execute_mcp_client - Ask a question to a separate LLM, ignore all the intermediate steps it takes when querying it’s tools, and return the output.
  • Takes question as required parameters
  • Returns answer, ignoring all the intermediate context
• execute_parallel_mcp_client - Takes a list of inputs and a main prompt, and executes the prompt in parallel for each string in the input. E.G. get the time of 6 major cities right now - London, Paris, Tokyo, Rio, New York, Sidney.
  • takes main prompt “What is the time in this city?”
  • takes list of inputs, London Paris etc
  • runs the prompt in parallel for each input
  • note: wait for this before using this feature
• execute_map_reduce_mcp_client - Process multiple items in parallel and then sequentially reduce the results to a single output.
  • Takes mapPrompt with {item} placeholder for individual item processing
  • Takes reducePrompt with {accumulator} and {result} placeholders for combining results
  • Takes list of items to process
  • Optional initialValue for the accumulator
  • Processes items in parallel, then sequentially reduces results
  • Example use case: Analyze multiple documents, then synthesize key insights from all documents into a summary

Development

Dependencies:

• Install mcp-client-cli
  • Also install the config file, and the mcp servers it needs in ~/.llm/config.json
• create a bash file somewhere that activates the venv and executes the llm executable

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

install package

Install dependencies:

npm install

Build the server:

npm run build

For development with auto-rebuild:

npm run watch

Installation

To use with Claude Desktop, add the server config:

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 from this repo
      "disabled": false,
      "autoApprove": [],
      "env": {
        "MCP_INCEPTION_EXECUTABLE": "./run_llm.sh", // bash file from Development->Dependencies
        "MCP_INCEPTION_WORKING_DIR": "/mcp-client-cli working dir"
      }
    }
  }
}

Debugging

Since MCP servers communicate over stdio, debugging can be challenging. We recommend using the MCP Inspector, which is available as a package script:

npm run inspector

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