MCP Server

Summary

All installed instances of NCrunch include an integrated MCP server that can be used by AI agents to query and steer the engine.

NCrunch's MCP server supports two operation modes:

• IDE Integrated Mode - Suitable for IDE hosted agents (such as GitHub Copilot and JetBrains AI Assistant)
• Headless Operation Mode - Suitable for fully autonomous agents (such as OpenClaw and Hermes)

IDE Integrated Mode

To enable NCrunch's MCP server inside your IDE, install the VS/Rider plugin as normal. Included with the installation files is NCrunch.MCP.exe, which should be adjacent to the core engine binaries. Before you can register the MCP server with your IDE, you will first need to locate the install path of NCrunch.MCP.exe.

In a VS-based installation, NCrunch's normal install path is under:
C:\Program Files\Microsoft Visual Studio\VERSION\SKU\Common7\IDE\Extensions\Remco Software\NCrunch for Visual Studio YEAR

Note that the root install path for VS can vary depending upon system configuration, but the relative path to the NCrunch install directory should be consistent.

The installation path of NCrunch under JetBrains Rider can vary greatly depending on how both the plugin and the IDE are installed. A normal expected path is:
C:\Users\USER\AppData\Roaming\JetBrains\RiderVERSION\plugins\ncrunchforrider\ncrunch

Depending on your configuration, it may be necessary for you to search your drive for the location of NCrunch.MCP.exe. Note that the NCrunch engine does make shadow copies of the engine while it is operating. Do not link the IDE to a copy of NCrunch.MCP.exe running inside a shadow copy sandbox, as these sandboxes can be removed when the engine shuts down.

Once you have located NCrunch.MCP.exe, you will need to add this as an MCP server in the IDE. In VS this is done through the tool settings in GitHub Copilot Chat. Under JetBrains Rider, you can register new MCP servers under Settings->Tools->AI Assistant->Model Context Protocol (MCP).

The NCrunch MCP server does not require any user-configured parameters or environment variables when running under IDE Integrated Mode.

Note that as of the time this documentation is written, IDE AI agents are undergoing heavy development and the steps required to set up an MCP server are in a state of flux. If you have any trouble adding the NCrunch MCP server, your AI agent should be able to help guide you through the process.

For the agent to be able to steer or make requests of NCrunch, the engine must be running, otherwise the agent will receive timeout errors from the MCP server. The MCP server can still operate while the engine is offline but no tool call will return a useful result.

Headless Operation Mode

Headless operation mode is an exciting feature that allows the NCrunch engine to run as a 24/7 service that can be used by fully autonomous agents. Agents are able to point the MCP server to any accessible solution on the machine, starting as many NCrunch engine sessions as necessary with full control of the lifespan of each session. Through NCrunch's filesync system, an agent can modify source code and have the engine automatically run tests and report results on request. In a manner of speaking, the NCrunch engine becomes a test-focused IDE allowing an agent to build and test their code while they work.

To run the MCP server in this way, you need to install the NCrunch Console Tool, which also includes NCrunch.MCP.exe.

Start NCrunch.MCP.exe with the parameters: /http address:port secret

For example:
NCrunch.MCP.exe /http 127.0.0.1:18088 afc8c7e3-aeb9-4ec9-b776-9b8268ca0331
This will start the MCP server, listening on port 18088 on the loopback network adapter, with a secret/password that must be provided by any connecting client to authenticate it.

IMPORTANT!!! Do not expose the MCP server to the open internet. Its integration with the NCrunch engine makes it a remote code execution tool and it is not designed to be a securely exposed service. Always restrict access to this service using appropriate firewall configuration.