MCP Inception

You can run a capable MCP server that delegates tasks to another MCP client, offloads context windows, and executes work in parallel or via map-reduce. This server acts as an agent for your agent, enabling nested querying and orchestration of LLM tools from a single interface.

How to use

Use this MCP server to call another MCP client from your own MCP client. You can delegate tasks, offload context windows, and run computations in parallel or in staged map-reduce fashion. Practical workflows include querying an external LLM for specialized tasks, aggregating results from multiple prompts, and synthesizing insights from several documents.

How to install

Prerequisites you need to prepare before running the server:

• Node.js and npm installed on your machine
• A working Python or shell environment if you plan to develop or run related tooling
• Access to the file paths referenced by your setup (see code snippet)

Install project dependencies and prepare the local environment as follows.

npm install

Build the server for production or local use.

npm run build

For development with auto-rebuild, start the watcher to rebuild on changes.

npm run watch

Additional setup and configuration

To integrate with Claude Desktop, configure the MCP server in the client’s MCP settings. The server is designed to run as a local stdio process that you start with a runtime command and a set of environment variables.

Example configuration for Claude Desktop (local stdio server):

{
  "mcpServers": {
    "mcp-inception": {
      "command": "node",
      "args": ["~/Documents/Cline/MCP/mcp-inception/build/index.js"],
      "disabled": false,
      "autoApprove": [],
      "env": {
        "MCP_INCEPTION_EXECUTABLE": "./run_llm.sh",
        "MCP_INCEPTION_WORKING_DIR": "/mcp-client-cli working dir"
      }
    }
  }
}

Debugging and troubleshooting

If you encounter issues, debugging is often easiest when you run the server in a controlled environment and inspect the standard I/O streams. Start by ensuring the runtime command and environment variables are accessible, then monitor the console for any startup errors or dependency failures.

You can also use a dedicated inspector to view debugging tools in your browser. Run the inspector script to get a URL for debugging utilities.

npm run inspector

Available tools

execute_mcp_client

Ask a question to a separate LLM, ignore intermediary steps, and return the final output.

execute_parallel_mcp_client

Run a main prompt in parallel across a list of inputs and collect results.

execute_map_reduce_mcp_client

Process multiple items in parallel and then reduce results sequentially to a single output.