> ## Documentation Index
> Fetch the complete documentation index at: https://densify-sync-changelog-51.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Tools: MCP Server

## Model Context Protocol

The Model Context Protocol, or MCP, is a standard that allows Large Language Models (LLMs) to securely access external systems,
tools, and data sources. By connecting through MCP, an LLM can retrieve real-time information and take structured actions,
extending its usefulness beyond static chat interactions. In practice, this means that an LLM can integrate live data
from trusted systems into its reasoning process, enabling more informed and actionable responses.

## Kubex MCP

Kubex MCP extends this capability to your cloud and container environments. It gives LLMs access to detailed operational
information about containers, nodes, and cloud instances, including Kubex optimization recommendations.
You can use a chat-based LLM interface to review key optimization insights, identify risks, and plan remediation
steps interactively. You can also use an LLM-based coding tool to apply Kubex recommendations directly to your
infrastructure-as-code definitions.

## Getting Started

### LLM Clients

To get started with Kubex MCP, you must use an LLM client that supports the Model Context Protocol.
You can choose an existing LLM client application or build your own. Kubex MCP should work with any LLM
client that supports MCP.

Common chat interfaces that support MCP include Anthropic Claude and ChatGPT when Developer Mode is enabled.
Coding-oriented clients such as Cursor, Claude Code, and Gemini CLI also provide MCP support.

Alternatively, you can develop your own LLM-based agent using MCP. This approach requires significant
development expertise and should only be attempted if you are comfortable building and maintaining your own
software. Frameworks such as Microsoft Copilot Studio and Pydantic AI provide a useful starting point for
developers who want to build their own MCP-enabled systems.

### Configuring Your Client

#### General

Each company has a unique Kubex MCP server endpoint. The MCP server URL follows a consistent pattern based on your company dashboard address.

* If your Kubex dashboard is [https://corp.kubex.ai](https://corp.kubex.ai/), your MCP server URL is [https://corp-mcp.kubex.ai](https://corp-mcp.kubex.ai/).
* If your organization uses a densify.com host such as [https://corp.densify.com/kubex](https://corp.densify.com/kubex), use the corresponding MCP URL: [https://corp-mcp.densify.com](https://corp-mcp.densify.com/).

Kubex MCP supports the streamable HTTP transport, using OAuth 2.1 with dynamic client registration for authentication.

#### Claude

##### Claude for Web (Claude.ai):

Navigate to the [connectors](https://claude.ai/customize/connectors) page and connect Kubex.

##### Claude Desktop:

From Claude Settings, go to Connectors and add the Kubex connector.

##### Claude Code:

Run the following command:

```bash theme={null}
claude mcp add --transport http kubex https://corp-mcp.kubex.ai/
```

Then run `/mcp` after you open a Claude Code session to sign in with your Kubex account and connect it to Claude Code.

#### Codex

The setup steps for the MCP server apply to both the IDE extension and the CLI because the configuration is shared.

Run the following command in your terminal:

```bash theme={null}
codex mcp add kubex --url https://corp-mcp.kubex.ai/
```

You will be prompted to sign in with your Kubex account and connect it to Codex.

#### Cursor

1. Press `Ctrl/Cmd + ,` to open settings.
2. Go to the **Tools & MCPs** tab.
3. Click **Add Custom MCP Server**.
4. Add the following configuration (replace the URL with your Kubex MCP URL):

```json theme={null}
{
	"mcpServers": {
		"kubex": {
			"command": "npx",
			"args": ["-y", "mcp-remote", "https://corp-mcp.kubex.ai/"]
		}
	}
}
```

You may need to restart Cursor for the MCP tools to appear in chat.

#### Visual Studio Code

1. Press `Shift + Ctrl/Cmd + P` and search for **MCP: Add Server**.
2. Select **HTTP**.
3. Enter your Kubex MCP URL and use `kubex` as the server name.
4. Authorize access to your Kubex instance when prompted.
5. Activate the server through **MCP: List Servers**, select `kubex`, then select **Start Server**.

You may need to restart Visual Studio Code for the MCP tools to appear in chat.

#### Devin Desktop (formerly Windsurf)

1. Press `Ctrl/Cmd + ,` to open settings.
2. Scroll to **Devin Local -> MCP servers**.
3. Select **Open Devin MCP marketplace -> Add custom MCP**.
4. Add the following configuration (replace the URL with your Kubex MCP URL):

```json theme={null}
{
	"mcpServers": {
		"kubex": {
			"command": "npx",
			"args": ["-y", "mcp-remote", "https://corp-mcp.kubex.ai/"]
		}
	}
}
```

Return to the Devin MCP marketplace and click **Connect** to authorize with Kubex.

#### Zed

1. Press `Ctrl/Cmd + ,` to open Zed settings.
2. Add the following configuration (replace the URL with your Kubex MCP URL):

```json theme={null}
{
	"context_servers": {
		"kubex": {
			"source": "custom",
			"command": "npx",
			"args": ["-y", "mcp-remote", "https://corp-mcp.kubex.ai/"],
			"env": {}
		}
	}
}
```

#### Others

Many other tools now support MCP servers. You can configure Kubex MCP with the following settings (replace the URL with your Kubex MCP URL):

* Command: `npx`
* Arguments: `-y mcp-remote https://corp-mcp.kubex.ai/`
* Environment: `None`

## Authorization

Access to Kubex data through MCP is tightly controlled. Only users who have been properly authenticated and who have explicitly authorized their LLM client application are allowed to retrieve or act on data. Authorization is temporary and must be refreshed daily for security reasons.

Some client applications handle expired authorizations gracefully, automatically prompting for reauthorization when required. Others may not, which can lead to connection errors or incomplete responses. If you encounter these issues, the solution is usually to manually reconnect and reauthorize your client with the Kubex MCP server.

## Security

Because MCP allows direct integration between an LLM and your operational environment, it is essential to treat security as a first priority. You should only connect to MCP servers that you trust. A malicious MCP server could attempt to manipulate an LLM into revealing sensitive information or performing unwanted actions.

Kubex guarantees that the Kubex MCP server is secure and trustworthy. It is designed to provide authorized users with access to relevant data and recommendations without compromising system integrity or user privacy. However, Kubex cannot make any guarantees about the behavior of third-party MCP servers. If you choose to use additional MCP connections, you do so at your own risk and should apply standard best practices for verifying the legitimacy and security of external systems.