Edit

Share via

MCP support

Model Context Protocol (MCP) is an open protocol that standardizes how applications provide context to large language models (LLMs). Starting in AI Shell 1.0.0-preview.6, AI Shell can act as an MCP Host and client to MCP servers. The key participants in the MCP architecture are:

  • MCP Host - AI Shell coordinates and manages one or multiple MCP clients
  • MCP Client - The client component of AI Shell maintains a connection to an MCP server and obtains context from an MCP server for the MCP host to use
  • MCP Server - A program, running locally or hosted remotely, that provides context to MCP clients

For more information about MCP, see the Architecture Overview in the official Model Context Protocol documentation.

MCP tools enable AI agents to access external tools and services to enhance their capabilities and provide more accurate responses. MCPs can integrate with various APIs, databases, and other resources, allowing agents to retrieve real-time information and perform complex tasks.

Built-in tools for AI Shell

AI Shell comes with a set of MCP-like built-in tools that are available to all agents. These commands are similar to MCP Server tools, but are exclusive to the AI Shell experience. The tools enhance the AI Shell experience by providing context-aware capabilities and automation features. You can use them in conjunction with other MCP servers to create a powerful AI-driven shell environment.

The following list contains the built-in tools and their usage:

  • get_working_directory - Get the current working directory of the connected PowerShell session, including the provider name (e.g., FileSystem, Certificate) and the path (e.g., C:\\, cert:\\).
  • get_command_history - Get up to 5 of the most recent commands executed in the connected PowerShell session.
  • get_terminal_content - Get all output currently displayed in the terminal window of the connected PowerShell session.
  • get_environment_variables - Get environment variables and their values from the connected PowerShell session. Values of potentially sensitive variables are redacted.
  • copy_text_to_clipboard - Copy the provided text or code to the system clipboard, making it available for pasting elsewhere.
  • post_code_to_terminal - Insert code into the prompt of the connected PowerShell session without executing it. The user can review and choose to run it manually by pressing Enter.
  • run_command_in_terminal - This tool allows you to execute shell commands in a persistent PowerShell session, preserving environment variables, working directory, and other context across multiple commands.
  • get_command_output - Get the output of a command previously started with run_command_in_terminal.

Note

The built-in tools rely on the side-car experience with a connected PowerShell session and provide enhanced context awareness and automation capabilities.

Demos

Here is a simple demo showing how you can have AI Shell run commands on your behalf using the run_command_in_terminal tool:

Have the MCP run a command in terminal for you.

This example shows how additional context is provided to AI Shell to improve results:

Getting more context with built-in tools.

You can also use the get_terminal_content tool to get the content from the connected terminal and provide it to AI Shell to help it understand what you are trying to do:

Getting content from the terminal for output created before AI Shell started.

Find and configure MCP servers

An MCP server is an application that runs locally or is hosted remotely and accessed through an endpoint URL. Many MCP servers are open source and available for free. Local MCP server apps are typically built using Node.js and can installed using npm. You can find a list of available MCP servers at Awesome MCP Servers.

Before you can use an MCP server (Local or remote) you must add it to the mcp.json configuration file. In some cases, you might not need to install the local MCP before using it. However, the MCP server will be installed the first time you start AI Shell. This can take several minutes. For the best experience, preinstall the MCP server. Consult the documentation for your specific MCP server for installation instructions. Remote MCP servers don't need to be installed. You just need to add them to the mcp.json file.

Install a local MCP server

The following example installs the @modelcontextprotocol/server-everything and @modelcontextprotocol/server-filesystem MCP servers. These servers are published as node packages. You must have Node.js installed to use these MCP servers.

To install the MCP servers, you can use the following commands:

npm install -g @modelcontextprotocol/server-everything
npm install -g @modelcontextprotocol/server-filesystem

For more information about their command and capabilities, see the following documentation:

Add MCP Servers

To add an MCP server, create an mcp.json file in $HOME\.aish\ folder. The following example shows two local MCP servers, everything and filesystem, and two remote MCP server. You can add as many MCP servers as you want.

{
    "servers": {
        "everything": {
            "type": "stdio",
            "command": "npx",
            "args": [
                "-y",
                "@modelcontextprotocol/server-everything"
            ]
        },
        "filesystem": {
            "type": "stdio",
            "command": "npx",
            "args": [
                "-y",
                "@modelcontextprotocol/server-filesystem",
                "C:/Users/username/"
            ]
        },
        "github": {
            "type": "http",
            "url": "https://api.githubcopilot.com/mcp/",
            "headers": {
                "Authorization": "Bearer <YOUR_GITHUB_TOKEN>"
            }
        },
        "microsoft.docs.mcp": {
            "type": "http",
            "url": "https://learn.microsoft.com/api/mcp"
        }
    }
}

For more information about the remote MCP server from Microsoft, see:

  • Last updated on