SCHEMAXIS · Declared work. Results you can check.Documentation ↗ · Contact
schemaxis/Agent’s Hands

Schemaxis for your computer

Your applications.
One repeatable workflow.

Download the free core, Schemaxis Studio and a local MCP adapter for your AI. Your native applications run on your computer; you choose what to run.

8.0 MiB · Python 3.10+ · Apache-2.0 · Core 0.4.0 · Local MCP adapter 0.1.0 (preview)

SHA-256 checksums · Build identity and file hashes

AVAILABLE

Linux

The supported platform. The bundle installs into a private Python environment without sudo or changing your PATH. Requires Python with venv and pip.

Install on Linux ↓
LINUX ON WINDOWS

Windows via WSL2

Install Ubuntu under WSL2, open its terminal and follow the Linux instructions. Use Linux builds of your native applications inside WSL. This path still needs verification on a Windows machine; it is not a native Windows installer.

In an administrator PowerShell, run wsl --install -d Ubuntu, restart if asked, then complete Ubuntu setup. If Python is missing in Ubuntu, install python3, python3-venv and python3-pip using Ubuntu’s package manager.

A Windows AI client must launch the installed command through wsl.exe; use the configuration below.

Microsoft’s WSL installation guide ↗
NOT YET QUALIFIED

macOS

No supported macOS download yet. Parts of the core use Unix interfaces, but process supervision and some application adaptors depend on Linux. macOS needs its own installation and native workflow checks.

01 / Install

Unzip. Install.
Choose your interface.

Extract the download and open a terminal in the extracted schemaxis-local folder. Run the installer below. It checks the bundled file hashes, creates a private environment and installs the included wheel without fetching Python packages.

The default folder is ~/.local/share/schemaxis. Use --directory /your/new/folder to choose another location. An existing folder is preserved; choose a fresh directory for an upgrade.

Native applications are installed separately. You do not need them to discover operations or plan an example.

python3 install.py

# Check the installed core
~/.local/share/schemaxis/venv/bin/schemaxis --version
~/.local/share/schemaxis/venv/bin/schemaxis discover

02 / Use Studio

Work visually,
in your browser.

Start Studio with this command and open the loopback URL printed in the terminal. Keep that terminal running while you work. Studio uses the same local core and declared application operations.

Studio user guide → · Configure native applications →

~/.local/share/schemaxis/venv/bin/schemaxis console \
  --web \
  --workspace ~/.local/share/schemaxis/studio-workspace

03 / Connect your AI

A free local MCP server.
Your AI starts it when needed.

The download includes a small stdio MCP adapter over the public core’s task, planning and evidence service. No Schemaxis account, server rental or API key is needed for this adapter. Your AI client’s subscription or model charges, and native application licences, are separate.

Codex

The installer writes codex-config.toml inside your installation folder. Copy that entry into your Codex MCP configuration, or use the command below, replacing /absolute/install with your actual installation folder.

codex mcp add schemaxis -- \
  /absolute/install/venv/bin/python \
  /absolute/install/schemaxis_mcp.py \
  --workspace /absolute/install/workspace

Official Codex MCP setup ↗

Other local MCP clients

The installer writes mcp-config.json with absolute paths. Add its schemaxis entry to a client that supports local stdio servers; merge it with any existing servers. It never changes your AI client’s settings automatically.

{
  "mcpServers": {
    "schemaxis": {
      "command": "/absolute/install/venv/bin/python",
      "args": ["/absolute/install/schemaxis_mcp.py",
        "--workspace", "/absolute/install/workspace"]
    }
  }
}
Using a Windows AI client with WSL2

Use wsl.exe as the command and prefix the Linux command with -d and the actual distribution name shown by wsl --list. Paths must be absolute Linux paths inside that distribution. Example arguments:

["-d", "Ubuntu", "--", "/absolute/linux/install/venv/bin/python",
 "/absolute/linux/install/schemaxis_mcp.py", "--workspace",
 "/absolute/linux/install/workspace"]

This launch configuration needs verification with your Windows client. Browser access to Studio from Windows also depends on WSL networking.

Planning mode, execution and permissions

The adapter starts in planning mode: discover contracts, create a task, plan and read evidence. To expose native execution, deliberately add --allow-execute to the server arguments and restart the connection. Configure the native application paths first and keep your AI client’s normal tool approvals enabled.

This trusted local adapter runs with your user account’s permissions. Task scopes and budgets are application controls, not an operating-system sandbox. It is distinct from the optional hosted service’s strict, isolated interface. The adapter supports MCP protocol 2025-11-25; a client must accept that negotiated version. A connection test is not a guarantee for every AI client or native application.

Upgrade or uninstall

Stop Studio and the MCP connection. Install a new bundle in a fresh directory and update your client’s entry to the new absolute paths. Preserve your workspace evidence and private application configuration. To uninstall, remove the MCP entry, preserve any results you want, then remove the installation directory. There is no system service, global PATH change or background startup registration to undo.

Your first workflow

Start with a plan.
Then run something small.

Follow the quick start →