# Mihu CLI: Manage Your AI Agents From the Terminal | Mihu AI Product update · Developers · September 2026 # Mihu CLI: Your Whole Workspace, One Command Away Some work belongs in a dashboard. Some belongs in a terminal, a script or a CI job. The Mihu CLI puts every part of your workspace, from AI agents and phone numbers to calls, conversations and campaigns, behind a single command, and connects your AI coding tools to Mihu in one line. Mihu AI Team September 9, 2026 6 min read v0.3.0 New · Mihu CLI Install · Log in · Run anything Your terminal One command Your whole workspace Everything you can do in the dashboard, you can now do from a terminal, a script or your AI coding tool. In this article 1 Why we built a CLI (https://mihu.ai/blog/mihu-cli#why) 2 Install it in one line (https://mihu.ai/blog/mihu-cli#install) 3 Log in once, switch workspaces freely (https://mihu.ai/blog/mihu-cli#login) 4 Every part of the API is a command (https://mihu.ai/blog/mihu-cli#commands) 5 Connect your AI coding tool in one command (https://mihu.ai/blog/mihu-cli#mcp) 6 Made for scripts and CI (https://mihu.ai/blog/mihu-cli#scripts) 7 Releases and updates (https://mihu.ai/blog/mihu-cli#releases) 8 Useful links (https://mihu.ai/blog/mihu-cli#links) 9 Frequently asked questions (https://mihu.ai/blog/mihu-cli#faq) ## Why We Built a CLI The dashboard is great for looking. A terminal is great for doing things again and again. The Mihu dashboard is where most people build and run their AI agents, and it will stay that way. But the teams using Mihu the most kept telling us the same thing. They wanted to create ten agents from one template, buy numbers for a new branch, check yesterday's calls before stand-up, or roll a change out from their deployment pipeline. Doing that with clicks is slow. Doing it with raw API requests means writing the same boilerplate every time. So we built the tool we wanted ourselves: one small program that knows your workspace, speaks the whole Mihu API, and gets out of your way. We had three goals for it. Nothing should be dashboard-only. If you can do it in Mihu, you can do it from a command. It should be friendly to people and to scripts. Readable tables when you are typing, clean JSON when a script is reading. Your AI tools should reach Mihu without setup pain. Connecting Claude Code, Cursor or VS Code to your workspace should take one line, not a config file. ## Install It in One Line One file, no runtime, macOS, Linux and Windows. The CLI is a single program with nothing else to install. The installer picks the right build for your computer, checks it has not been tampered with, and puts it on your path. Terminal # macOS and Linux curl -sSL https://mihu.ai/install.sh | bash # Windows (PowerShell) irm https://mihu.ai/install.ps1 | iex mihu version That is the whole setup. It works on Intel and ARM machines alike. ## Log In Once, Switch Workspaces Freely Your token is saved safely, and you never type it again. Run mihu login . It asks for your workspace, opens the page where you create an API token, and saves the token on your computer with permissions only you can read. From then on every command knows who you are. mihu whoami shows which workspace you are in and your balance. Working with more than one workspace, such as production and staging, or several clients? Give each login a name and switch between them with one flag. On a build server, set two environment variables instead and nothing is stored on disk at all. Terminal mihu login --name production mihu login --name staging mihu -p staging agents list mihu config use production ## Every Part of the API Is a Command If the API can do it, the CLI can do it, and it follows the same shape. We did not hand-pick a few popular actions. Every endpoint of the Mihu API is a command, and the commands follow the same structure as the API, so once you know one area you can guess the rest. mihu commands lists all of them. Area What you can do AI agents Create, update and delete agents, build one from a website, manage intents, voice, webhooks and channels, restore earlier versions and answer what the agent asks for Calls Start a call, list and read calls, and act on a live call: say something, mute, forward or hang up Numbers Search, buy, bind and release phone numbers Conversations List, read and reply to conversations on every channel, and change their status Contacts & campaigns Manage contacts, tags and fields, and create and publish campaigns Tickets, quality, knowledge Tickets and their rules, evaluations and analytics, knowledge bases and data tables You do not have to remember every field either. Run a create or update command with nothing after it and the CLI asks you for each value in turn, explains what it is, offers the allowed options, and shows you the request before it sends anything. Anything the CLI has no dedicated command for is still one line away with mihu api , which sends a signed-in request to any address in the API. Terminal mihu agents create --name "Front desk" --language en --tone "Short and friendly" mihu calls start +14155550123 --agent --prompt "Confirm tomorrow's appointment" mihu conversations list --channel whatsapp --status open mihu api /workspace Log in, list your AI agents, put one on a call and connect Claude Code, all in about half a minute. ## Connect Your AI Coding Tool in One Command Let Claude Code, Cursor or VS Code manage your workspace in plain language. The Mihu MCP server lets AI clients work with your workspace by talking to them: list my agents and tell me which ones have no phone number . The hard part used to be the setup: finding the right address, the right headers and the right config file for each tool. The CLI does all of that with the workspace and token you already logged in with. Terminal mihu mcp setup claude-code mihu mcp setup cursor mihu mcp setup vscode mihu mcp setup windsurf mihu mcp setup claude mihu mcp status One line per tool. Restart the tool and ask it something. The MCP server offers several hundred tools, and AI clients work better with fewer. You can narrow them to one area, such as the inbox, appointments, campaigns or building agents. When you no longer need the connection, mihu mcp remove takes it out again. Step-by-step setup for every tool is in the MCP guide (https://developers.mihu.ai/cli/mcp) . ## Made for Scripts and CI Readable for you, predictable for your scripts. ### Tables or JSON, automatically You see a readable table when you type a command. When its output goes into another program, you get JSON instead, ready for tools like jq. ### Every page at once Lists page through the results for you when you ask for all of them, so a report or an export is one command. ### No prompts in pipelines On a build server, a token comes from environment variables, and destructive commands can be confirmed up front so nothing waits for a keypress. ### Safe by default Deleting something or releasing a number always asks first when a person is at the keyboard. Put together, this means a deployment pipeline can name the current version of an agent, apply a change, run your tests, and roll the agent back (https://mihu.ai/blog/agent-versions) if they fail, all without anyone opening the dashboard. ## Releases and Updates It keeps itself current, and every release is public. mihu update installs the newest version, and the CLI quietly tells you when one is available. Shell completions for bash, zsh, fish and PowerShell are one command away, and mihu uninstall removes everything cleanly. Every release is published with checksums on GitHub (https://github.com/mihu-ai/mihu-cli-releases) . Version Date What changed v0.1.0 September 4, 2026 First release: every API endpoint as a command, profiles, one-line MCP setup v0.2.0 September 7, 2026 Tickets, with their statuses, tags, rules and notes v0.2.1 September 7, 2026 The update check no longer slows a command down v0.3.0 September 18, 2026 Agent versions and Auto-Improvement requests: list, restore, answer, skip and automate Try this: install the CLI, run mihu login , then mihu mcp setup for the AI tool you already use. Ask it to list your agents and check which ones still need a phone number. ## Useful Links CLI guide (https://developers.mihu.ai/cli/overview) : install, log in and every command. MCP guide (https://developers.mihu.ai/cli/mcp) : connect Claude Code, Claude Desktop, Cursor, VS Code and Windsurf. Releases on GitHub (https://github.com/mihu-ai/mihu-cli-releases) : every version with its checksums. ## Frequently Asked Questions Do I need to install anything else first? No. The CLI is a single program with no other requirements. The only exception is connecting Claude Desktop, which uses a small bridge that needs Node.js. Where is my token stored? In a file in your user folder that only your account can read. On a build server you can use environment variables instead, so nothing is stored. Logging out removes it. Can the CLI do everything the dashboard can? Everything the Mihu API can do, yes: every endpoint is a command. Anything without its own command can still be reached with mihu api. Can I use it with more than one workspace? Yes. Log in once per workspace with a name, then switch with a flag or make one the default. Which AI tools can it connect to Mihu? Claude Code, Claude Desktop, Cursor, VS Code and Windsurf in one command each. For any other MCP client, mihu mcp url prints the connection details. How do I keep it up to date? Run mihu update. The CLI also tells you when a new version is available. ## Install the Mihu CLI One line to install, one command to log in, and your whole workspace is in your terminal. The full guide is in the developer documentation. Read the CLI guide → (https://developers.mihu.ai/cli/overview) Product update Developers CLI MCP Automation ## Keep exploring More ways to build with Mihu AI 🔌 Product #### The Mihu MCP Server Blog (https://mihu.ai/blog/mcp-server) 📋 Product update #### Agent Versions: We Save Every Change to Your Agent Sep 2026 (https://mihu.ai/blog/agent-versions) 💻 Developers #### API & Developers Product page (https://mihu.ai/api-reference)