Getting started
Install INXM Local
Install a release package when you want to use the desktop application.
Quick install
The recommended way to install INXM Local is to use the quick install command for your operating system. It downloads the latest release and installs it per-user (no root needed).
๐ macOS
curl -fsSL https://raw.githubusercontent.com/inxm-ai/inxm-local/main/packaging/install.sh | sh
๐ช Windows
irm https://raw.githubusercontent.com/inxm-ai/inxm-local/main/packaging/install.ps1 | iex
๐ง Linux
curl -fsSL https://raw.githubusercontent.com/inxm-ai/inxm-local/main/packaging/install.sh | sh
The scripts install per user. The Unix installer supports --version, --autostart, --agents, individual agent flags, and --uninstall. Existing agent configuration is merged and registration is idempotent. Use INXM_MCP_URL when registering an endpoint other than the default.
Manual installation
Download a package from GitHub Releases:
๐ macOS
Choose aarch64 for Apple Silicon or x86_64 for Intel, unzip the .app.zip, and open the app.
Release builds are not notarized or signed. macOS may require Open Anyway in System Settings -> Privacy & Security, or removal of the quarantine attribute after unzipping:
xattr -dr com.apple.quarantine ~/Applications/"INXM Local.app"
๐ช Windows
Run the x86_64-pc-windows-msvc-setup.exe installer.
Windows SmartScreen may require More info -> Run anyway or:
Unblock-File .\inxm-local-x86_64-pc-windows-msvc-setup.exe
๐ง Linux
- Debian or Ubuntu: install the
x86_64-unknown-linux-gnu.debpackage withsudo apt install ./<package>.deb. - Without root: use the release
.tar.gzand itsinstall.shscript.
Other installer options
| macOS/Linux | Windows | What it does |
|---|---|---|
--agents |
-Agents |
Register every supported agent detected on the machine. |
| Individual agent flags | Matching switches | Register one or more specific coding agents; see the table below. |
--autostart |
Not available | Linux: start INXM Local hidden at login. |
--version 0.1.0 |
-Version 0.1.0 |
Pin the installation to a specific release. |
--uninstall |
Windows app uninstall | Remove the app and its agent registrations. |
Installer environment variables
| Variable | What it does |
|---|---|
INXM_MCP_URL |
MCP endpoint registered with coding agents; defaults to http://127.0.0.1:39387/mcp. |
PREFIX |
Linux installation root; defaults to $HOME/.local. |
Both installer scripts are attached to each release and can be downloaded directly from the latest release.
Verify the installation
Open INXM Local. On a new install, complete or skip the first-run assistant to reach the chat view. Select a compiler connection under Settings -> Compiler before compiling a plan. If the local MCP server is enabled, its status and port are shown in the application settings.
Register coding agents
The standard installers above install INXM Local only. Connecting a coding agent such as Claude Code or Codex is an optional way to use INXM Local: the agent can create and manage reusable workflow plans for you through the local MCP server.
๐ macOS
The macOS installer is a shell script. To register every supported agent already detected on your machine, add --agents:
curl -fsSL https://raw.githubusercontent.com/inxm-ai/inxm-local/main/packaging/install.sh | sh -s -- --agents
To register one specific agent, pass its flag instead, for example --claude:
curl -fsSL https://raw.githubusercontent.com/inxm-ai/inxm-local/main/packaging/install.sh | sh -s -- --claude
Refer to the Supported agents section for more details on the available flags and switches.
๐ช Windows
Download install.ps1, then run it in PowerShell. To register every supported agent already detected on your machine, use -Agents:
.\install.ps1 -Agents
To register one specific agent, use its switch instead, for example -Claude:
.\install.ps1 -Claude
Refer to the Supported agents section for more details on the available flags and switches.
๐ง Linux
The Linux installer uses the same shell commands as macOS. To register every supported agent already detected on your machine, add --agents:
curl -fsSL https://raw.githubusercontent.com/inxm-ai/inxm-local/main/packaging/install.sh | sh -s -- --agents
To register one specific agent, pass its flag instead, for example --claude:
curl -fsSL https://raw.githubusercontent.com/inxm-ai/inxm-local/main/packaging/install.sh | sh -s -- --claude
Refer to the Supported agents section for more details on the available flags and switches.
Supported agents
The Unix flags and Windows switches below perform the same registration. Existing configuration files are merged rather than overwritten, and registration is idempotent: rerunning the installer does not duplicate entries.
| macOS/Linux flag | Windows switch | Agent | Registration |
|---|---|---|---|
--claude |
-Claude |
Claude Code | Runs claude mcp add at user scope and installs the use-inxm-mcp skill. |
--codex |
-Codex |
Codex CLI | Adds the MCP server to the Codex configuration. |
--gemini |
-Gemini |
Gemini CLI | Adds the MCP server to the Gemini settings. |
--qwen |
-Qwen |
Qwen Code | Adds the MCP server to the Qwen settings. |
--copilot |
-Copilot |
GitHub Copilot CLI | Adds the MCP server to the Copilot configuration. |
--vscode |
-VSCode |
VS Code (Copilot) | Adds the MCP server to the user-level mcp.json. |
--cursor |
-Cursor |
Cursor | Adds the MCP server to the Cursor configuration. |
--windsurf |
-Windsurf |
Windsurf | Adds the MCP server to the Windsurf configuration. |
--cline |
-Cline |
Cline | Adds the MCP server to the Cline settings. |
--roo |
-Roo |
Roo Code | Adds the MCP server to the Roo Code settings. |
--opencode |
-OpenCode |
OpenCode | Adds the MCP server to the OpenCode configuration. |
--goose |
-Goose |
Goose | Adds the MCP server to the Goose configuration. |
--hermes |
-Hermes |
Hermes | Runs hermes mcp add; see the Hermes integration guide. |
--pi |
-Pi |
Pi | Installs the use-inxm-mcp skill; Pi has no native MCP configuration. |
--zed |
-Zed |
Zed | Adds the MCP server to Zed and installs the use-inxm-mcp skill. |
See Use INXM Local with coding agents for the complete guide.
What's next
- Create your first workflow and turn a natural-language request into a saved plan.
- Connect coding agents to enable the agent to create and manage reusable workflow plans.