1. Kion Success Center
  2. AI
  3. MCP Client

MCP Server Installation and Configuration

Follow

Last updated

Overview

Model Context Protocol (MCP) is a standardized way for tools, such as Kion, to interface with Large Language Models (LLMs) such as Claude as a plug-in. Kion provides an MCP server implementation that runs on an end-user's desktop system and interfaces with a Kion instance. This article provides instructions for any end-user of Kion to configure the Kion MCP server and make use of it with an LLM. 

Requirements

The following requirements must be met in order to use the Kion MCP Server:

  • Your target instance of Kion must be running v3.12.x or higher. Older versions of Kion are not supported with the MCP server even though some information may be returned.
  • You must have permission to create an API key in your Kion instance and you must have permission to perform any of the desired operations through the MCP server.
  • You must be able to install these prerequisites on your workstation:
    • uv (via Brew or Chocolatey as described below)
    • Python (Windows only; MacOS includes Python by default)

Setup Instruction

Follow the steps in this section to install the Kion MCP server on your local system. 

Before installing the Kion MCP you will need to have uv (a python package manager) installed. Follow these steps based on your system:

  1. For MacOS:
    1. Ensure that you have Homebrew installed using this command: /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
    2. Install the uv package: brew install uv
  2. For Windows:
    1. Ensure that you have Chocolatey installed. Use these steps as a non-administrator.
    2. Install the uv package: choco install uv
    3. For Claude, you must also have Python installed on Windows. If you don't already have this installed and configured, you can download and install the package from the Python website here.
      1. NOTE: When installing, ensure to check the option Add python.exe to PATH. Otherwise, Claude may fail to detect Python.

NOTE: If the Kion MCP has an error during startup, it’s typically caused by uv not being properly installed and added to path. Repeating the uv installation steps with one of the recommended package managers and then restarting your MCP client usually fixes this.

To finish installing the Kion MCP you can either have a guided installation through Claude Desktop or a manual installation which works with any MCP Client:

Claude Desktop (recommended)

  1. Download and install Claude Desktop from this link.
  2. Install Claude and ensure that you completely sign-in and answer all of the available wizards (prompts for Claude Code, etc - you can decline them but they must be answered). You are ready to proceed when you are at the How can I help you? prompt in Claude.
  3. Download the Kion MCP DXT file, located in release section of Kion's GitHub Repo.
  4. Double-click/open the DXT file you downloaded in the previous step.
    1. On MacOS, this will open in Claude automatically.
    2. On Windows, follow these steps to associate the DXT file to Claude:
      1. You will be prompted with Select an app to open this .dxt file. Select Choose an app on your PC.
      2. An Open With dialog will appear. In the path at the top, paste %LOCALAPPDATA%\AnthropicClaude and press Enter.
      3. In the listing of files, select claude and choose Open.
      4. You will be returned to the app selection screen. Choose Just once to continue. This will open Claude.
  5. You should see an installation preview appear in Claude Desktop. Click install and repeat this if prompted to do so again.
    1. NOTE: If you do not see the prompt titled Kion MCP Server, you may not have had Claude Desktop running. Just re-open the DXT file again using Claude.
    2. On Windows: If the install button is disabled and Claude reports that this extension is not compatible with your device, this may be because Python is not properly detected. Follow these steps:
      1. Fully exit Claude. This means both closing the window and locating the tray icon, right-clicking, and choosing Quit.
      2. With Claude closed, open PowerShell and type python -V. You should see a version print out at the command prompt. If you are taken to the Windows store or receive an error about opening a Windows Store link, you do not have Python in your PATH. Please add Python to your PATH variable and try again.
  6. A configuration window should pop up asking for the URL of your Kion instance and the Bearer Token. Follow its instructions to retrieve and enter these values.
    1. NOTE: The Bearer Token field should be populated with only the API key provided by Kion. Do not place a prefix on this value such as Bearer - use the token only.
  7. You should now see the same preview window as in step 5, but now with a toggle that shows Disabled. Toggle this selection so that it now shows Enabled.
  8. You’re done! Close out of the preview window and the Kion MCP should now be available for Claude to use.
    1. NOTE: The first time you execute a query, you will be prompted about using the Kion MCP Server. Select Allow always to allow Claude to use this integration without prompting in the future.
    2. NOTE: On the Windows version, you may appear to be stuck in Settings. Click the orange (+) New Chat button on the left-hand side to return to the main prompt.

Manual Setup

  1. Download the kion-mcp.zip file, located in release section of Kion's GitHub Repo
  2. Download the kion_mcp_config.yaml file and move it into your user directory
  3. Edit the kion_mcp_config.yaml file to contain your Kion instance url and a user App API Key. You can create a user App API Key in your Kion user menu in the top right of Kion by selecting “App API Keys” from that dropdown and then clicking “Add +”
  4. Unzip the kion-mcp.zip file and navigate to your MCP Client’s MCP Server json file

  5. Add the following object to that file:

      "KionMcp": {
     "command": "uv",
     "args": [
       "--directory",
       "<full path to your kion-mcp folder>",
       "run",
       "kion-mcp-server"
     ]
   }
  6. Alternatively if your MCP Client lets you paste in the command to run the server just paste this with the updated path to the unzipped Kion MCP folder: uv --directory <full_path_to_kion-mcp_folder> run kion-mcp-server
  7. Restart your MCP Client and the Kion MCP should now be available!

FAQs

How do I get started with setup?

Please refer to Kion’s setup instructions for the necessary steps for accessing the Kion MCP. For all other setup questions specific to the client or questions about how they work, please refer to the client specific documentation (Anthropic example).

What can I use the Kion MCP for?

There are defined set of tools (API endpoints) the MCP can access within Kion. These are all actions a user can take within Kion today via our public API, but these tools embedded into a MCP allow for efficiency and ease of use. To get the best experience, we recommend using a general AI client like Claude. At a high level, below are the actions you can take:

Organization Management

  • Create and manage Organizational Units (OUs) to structure your cloud environment
  • Set up projects within OUs
  • View and filter accounts, OUs, and projects by various criteria including labels

Financial Management and Reporting

  • Create funding sources to allocate budgets across your organization
  • Set up budgets for projects and OUs with detailed spending plans
  • Generate comprehensive spend reports with multiple filtering options (by OU, project, account, cloud provider, services, etc.)
  • Generate detailed spend analysis with various dimensions (services, regions, resources, etc.)
  • Filter and analyze data across your entire cloud environment

Compliance & Security

  • Create and manage compliance checks using Cloud Custodian policies or Azure Policies
  • Monitor compliance findings and suppressed findings across your environment
  • Track compliance against various standards and frameworks
  • View compliance status at the OU, project, and account levels
  • Create compliance reports and track findings
  • Apply permission schemes to OUs, projects, and funding sources
  • Manage ownership through user and group assignments

Explore our AI Prompt Library for ready-to-use example prompts designed for both our MCP and Lux, our in-app AI agent. Some prompts rely on additional MCP integrations such as Pendo, GitHub, GitLab, AWS, Azure, Google Cloud, Slack, and more, to enhance reporting and analysis with richer, more comprehensive data, as well as to connect to and take action across systems. With the MCP server, Kion seamlessly integrates into your broader AI ecosystem.

What can’t it do?

The tools cannot take any action that Kion hasn’t given explicit access to, and the above examples cover the feature areas currently permitted. Additionally, there are currently no DELETE endpoints approved, meaning these tools cannot delete/remove any item inside of Kion.

How can I block users from utilizing a MCP?

As Kion does not deploy with a MCP server, an organization’s policies or controls for installing 3rd party software on user’s devices applies. Users must be able to run the MCP server on their device.

Are there any actions within Kion I can take once a MCP is on a local desktop?

The Kion MCP server relies on an end user’s permissions. Kion administrators should carefully review user permissions and API key configurations to ensure they align with organizational policies.

I receive the message 'The MCP server appears to need an update.' or similar. How should I handle this?

The likely cause of this error message is that the LLM you're using is experiencing errors with the MCP server while trying to interact with Kion. You should ensure that your Kion instance is running at least v3.12.x and is accessible from the same system where you're running your MCP server.

How do I refresh the configuration of the MCP server (server URL, server token)?

The simple answer is to restart the MCP server (this may be handled by your LLM). With Claude Desktop, for example, you just restart the Claude Desktop application. This will both refresh Claude and restart the Kion MCP Server if you used the configuration steps in this document.

My organization doesn't permit the use of Claude Desktop. How can I proceed?

Claude Desktop is what we've chosen to use as an example with the MCP server we provide but it's certainly not your only choice. Any LLM that supports MCP for external tools can make use of the Kion MCP Server giving you the choice to explore options that meet your organization's requirements. We provide instructions on how to setup the MCP server for use outside of Claude (using the steps in the Manual Setup section above) and you would just reference to your chosen LLM's setup instructions on how to connect it once configured.

Reminder - AI Can Be Helpful, But Not Always Accurate

Kion is exploring the integration of AI-powered features to help streamline your cloud governance experience. These tools are designed to enhance productivity by providing helpful suggestions, summaries, and insights. However, it's important to be aware of the limitations of AI-generated content.

AI tools may sometimes generate content that is incorrect, incomplete, or misleading (commonly referred to as hallucination). This means the AI might present information that sounds plausible but is not factually accurate.

Using these tools with Kion:

  • Verify Before Acting: Always double-check any recommendations or summaries, especially when making important decisions about cloud configurations, cost management, or compliance actions.
  • Use AI as a Guide, Not a Final Answer: Think of AI tools as helpful assistants for Kion actions. They can speed up tasks and point you in the right direction, but they shouldn’t replace your own expertise or established workflows.
  • Report Concerns: If you encounter AI-generated content that seems incorrect or confusing, please let us know so we can continue improving the experience.

At Kion, we are committed to empowering our users with advanced tools in a responsible way. Your trust and security remain our top priorities.