- \n
- MCP client integration<\/li>\n
- Built-in tools<\/li>\n
Resolve-Error<\/code> command improvements<\/li>\n- Aliases and flows for staying in your terminal<\/li>\n<\/ul>\n
MCP Integration<\/h2>\n
AI Shell now acts as an MCP client, which allows you to add any MCP server to your AI Shell\nexperience. Connecting to an MCP server massively improves the capability of your AI Shell giving\nyou the tools that provide more relevant data or carry out actions!<\/p>\n
![\"AI](\"https:\/\/devblogs.microsoft.com\/powershell\/wp-content\/uploads\/sites\/30\/2025\/08\/AISHMCPDemo.gif\")
<\/p>\nAdding MCP Servers<\/h3>\n
To add an MCP server, create an
mcp.json<\/code> file in$HOME\\.aish\\<\/code> folder. The following example\nshows two MCP servers:everything<\/code> andfilesystem<\/code>. You can add any MCP servers you want.<\/p>\n{\r\n \"servers\": {\r\n \"everything\":{\r\n \"type\":\"stdio\",\r\n \"command\":\"npx\",\r\n \"args\":[\"-y\", \"@modelcontextprotocol\/server-everything\"]\r\n },\r\n \"filesystem\": {\r\n \"type\": \"stdio\",\r\n \"command\": \"npx\",\r\n \"args\": [\r\n \"-y\",\r\n \"@modelcontextprotocol\/server-filesystem\",\r\n \"C:\/Users\/username\/\"\r\n ]\r\n }\r\n }\r\n }<\/code><\/pre>\nIf it’s a remote MCP server, change the type to
https<\/code>. You know that you have successfully added\nan MCP server when you see it in the AI Shell UI. You can confirm that it’s running by checking the\nstatus of the server through the\/mcp<\/code> command. Using\/mcp<\/code> also lists each MCP Server and the\ntools available.<\/p>\n![\"Screenshot](\"https:\/\/devblogs.microsoft.com\/powershell\/wp-content\/uploads\/sites\/30\/2025\/08\/mcpui.jpeg\")
<\/p>\n<\/i>NOTE<\/strong><\/p>You must have
Node.js<\/code> oruv<\/code> installed to use MCP servers that\nuse those command lines tools.<\/div><\/p>\nStandalone experience with AI Shell and MCP Servers<\/h3>\n
MCP servers enhance your standalone experience with AI Shell, allowing your command line to use MCP\nservers and AI to perform tasks. For example,
@simonb97\/server-win-cli<\/code><\/a> is an MCP server that\nallows you to run commands on your Windows machine, whether it be PowerShell, CMD, Git Bash, or any\nconfigured shell you use! It also provides configuration settings to define which commands and\noperations are allowed to run.<\/p>\n<\/i>CAUTION<\/strong><\/p>Please note this is a community MCP server and not an\nofficial Microsoft MCP Server. We encourage you to do your own research and testing before using\nit.<\/div><\/p>\n
![\"AI](\"https:\/\/devblogs.microsoft.com\/powershell\/wp-content\/uploads\/sites\/30\/2025\/08\/standaloneMCPDemo.gif\")
<\/p>\nAdditional MCP servers:<\/p>\n
- \n
- DesktopCommander<\/a><\/li>\n
- Other community MCP Servers<\/a><\/li>\n<\/ul>\n
Built-in Tools for AI Shell<\/h2>\n
This release introduces built-in tools that are now accessible to agents within AI Shell. These\ncommands are similar to MCP Server tools, but are exclusive to the AI Shell experience. These tools\nare designed to enhance the AI Shell experience by providing context-aware capabilities and\nautomation features. They can be used in conjunction with the MCP servers to create a powerful\nAI-driven shell environment.<\/p>\n
\n\n
\n \nTool Name<\/th>\n Description<\/th>\n<\/tr>\n<\/thead>\n \n get_working_directory<\/code><\/td>\nGet the current working directory of the connected PowerShell session, including the provider name (e.g., FileSystem<\/code>,Certificate<\/code>) and the path (e.g.,C:\\\\<\/code>,cert:\\\\<\/code>).<\/td>\n<\/tr>\n\n get_command_history<\/code><\/td>\nGet up to 5 of the most recent commands executed in the connected PowerShell session.<\/td>\n<\/tr>\n \n get_terminal_content<\/code><\/td>\nGet all output currently displayed in the terminal window of the connected PowerShell session.<\/td>\n<\/tr>\n \n get_environment_variables<\/code><\/td>\nGet environment variables and their values from the connected PowerShell session. Values of potentially sensitive variables are redacted.<\/td>\n<\/tr>\n \n copy_text_to_clipboard<\/code><\/td>\nCopy the provided text or code to the system clipboard, making it available for pasting elsewhere.<\/td>\n<\/tr>\n \n post_code_to_terminal<\/code><\/td>\nInsert 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.<\/td>\n<\/tr>\n \n run_command_in_terminal<\/code><\/td>\nThis tool allows you to execute shell commands in a persistent PowerShell session, preserving environment variables, working directory, and other context across multiple commands.<\/td>\n<\/tr>\n \n get_command_output<\/code><\/td>\nGet the output of a command previously started with run_command_in_terminal<\/code>.<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/i>Note<\/strong><\/p>The built-in tools rely on the side-car experience with a\nconnected PowerShell session and provide enhanced context awareness and automation capabilities.<\/div><\/p>\n
Here is a simple demo showing how you can have AI Shell run commands on your behalf using the\n
run_command_in_terminal<\/code> tool:<\/p>\n![\"Run](\"https:\/\/devblogs.microsoft.com\/powershell\/wp-content\/uploads\/sites\/30\/2025\/08\/openai-agent-1.gif\")
<\/p>\nThis example shows how additional context is provided to AI Shell to improve results:<\/p>\n
![\"Getting](\"https:\/\/devblogs.microsoft.com\/powershell\/wp-content\/uploads\/sites\/30\/2025\/08\/openai-agent-2.gif\")
<\/p>\nYou can also use the
get_terminal_content<\/code> tool to get the content from the connected terminal and\nprovide it to AI Shell to help it understand what you are trying to do:<\/p>\n![\"Getting](\"https:\/\/devblogs.microsoft.com\/powershell\/wp-content\/uploads\/sites\/30\/2025\/08\/openai-agent-3.gif\")
<\/p>\nResolve-Error Command Improvements<\/h2>\n
Previously the
Resolve-Error<\/code> command was only able to run after an error occurred in the previous\ncommand. Now,Resolve-Error<\/code> identifies which command the user wants to troubleshoot:<\/p>\n- \n
- If the last error’s command matches the most recent command in history, it’s assumed to be the one\nthe user is interested in.<\/li>\n
- If the last error’s command isn’t the most recent and
$LastErrorCode<\/code> is null or zero, the error\nlikely comes from an earlier command, not the very last one.<\/li>\n- If
$LastErrorCode<\/code> is non-zero and$?<\/code> is false, the last command was a failing native command.<\/li>\n- If
$LastErrorCode<\/code> is non-zero but$?<\/code> is true, it’s unclear which command or failure the user\nis focused on, so the agent analyzes the terminal content to determine the relevant context.<\/li>\n<\/ul>\nThis logic allows AI Shell to better understand what the error the user is trying to resolve is\nrather than requiring you to ask for AI’s help immediately after an error occurs.<\/p>\n
Staying in your shell<\/h2>\n
The
Invoke-AIShell<\/code> andResolve-Error<\/code> commands allow you to stay in your working terminal to\ninteract with the AI Shell agent. To learn more about the parameters added, see the\nprevious blog post<\/a> that details these features. For your convenience, these commands have\naliases that make them quicker to use.<\/p>\n\n\n
\n \nCommand Name<\/th>\n Alias<\/th>\n<\/tr>\n<\/thead>\n \n Invoke-AIShell<\/code><\/td>\naskai<\/code><\/td>\n<\/tr>\n\n Resolve-Error<\/code><\/td>\nfixit<\/code><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n![\"Fixing](\"https:\/\/devblogs.microsoft.com\/powershell\/wp-content\/uploads\/sites\/30\/2025\/08\/openai-agent-4.gif\")
<\/p>\nHow to install AI Shell<\/h2>\n
To install the latest version of AI Shell, run the following command in your PowerShell terminal:<\/p>\n
Invoke-Expression \"& { $(Invoke-RestMethod 'https:\/\/aka.ms\/install-aishell.ps1') }\"<\/code><\/pre>\n - If
- Other community MCP Servers<\/a><\/li>\n<\/ul>\n
- Aliases and flows for staying in your terminal<\/li>\n<\/ul>\n