Connect an MCP host

The CLI ships an MCP server. Point any MCP host at it and your agent gets StratoNext as a set of typed tools: no commands to memorise, no output to parse, and no terminal in the loop.

Note

The MCP server exposes a subset of the platform: the tools needed to carry out work. For the full surface, use the CLI with the agent skill, which covers every command including organizations, workspaces, members, agents, jobs, and deployments.

Pick MCP when the agent’s job is to do the work: pick up a task, act on it, write up the result. Pick the CLI and skill when the agent also needs to set things up or administer the platform, or when it cannot speak MCP.

Prerequisites

Install the CLI and log in:

uv tool install stratonext
stratonext login

The MCP server reuses that login, so there is nothing separate to authenticate.

Add the server

For Claude Code:

claude mcp add stratonext -- stratonext mcp

For Claude Desktop, Cursor, or any other host, add this to its MCP configuration file:

{
  "mcpServers": {
    "stratonext": {
      "command": "stratonext",
      "args": ["mcp"]
    }
  }
}

Restart the host and ask your agent something like “what StratoNext tasks are ready for me?” to confirm it is connected.

Note

The server cannot open a browser, so it will never log you in by itself. If you are not signed in, every tool answers with a reminder to run stratonext login first.

What your agent can do

The tool set is deliberately small. It covers carrying out work, not administering the platform:

• Projects: list them and read full detail, including linked environments and deployments.
• Tasks: list, read, create, update, move through their lifecycle, and answer a task that is waiting on input.
• Reports: list, read, and create write-ups that outlive a single task.
• Environments: list and read the environments connected to the workspace.
• Assets: upload and download files attached to a project, task, or report.
• Approvals: see what is gating a task.

Everything else the CLI can do is left out on purpose, because a tool the agent never needs still costs it attention on every request. Organizations, workspaces, members, users, agents, scheduled jobs, deployments, and deletions are not exposed here. Use the web app for those, or the CLI and agent skill if you want an agent to handle them.

What it will not do

Two limits are deliberate, and your agent is told about both:

• It cannot approve anything. Approving or rejecting is a human decision, made in the web app. There is no tool for it, in any mode.
• It will not hand you cloud credentials. In a normal signed-in session StratoNext gives your agent context, not credentials: it works under your own cloud credentials, and it will not assume a role or find a way around a permission it lacks. If the work needs access you do not have, the agent creates a task for a background agent instead.

The second limit does not change for a person. Only an unattended process running the server as an agent can draw short-lived environment credentials from the credential broker. Do not run your own session as an agent to get around it.

Running it unattended

The same server can run as an agent. Set the agent’s credential in the host’s environment block and it runs as that agent rather than as you:

{
  "mcpServers": {
    "stratonext": {
      "command": "stratonext",
      "args": ["mcp"],
      "env": {
        "STRATONEXT_AUTH_API_CLIENT_ID": "agtc_...",
        "STRATONEXT_AUTH_API_CLIENT_SECRET": "..."
      }
    }
  }
}

See Operating modes for when that is the right choice.