Hyperglance MCP

Overview

MCP (Model Context Protocol) is an open standard that lets AI assistants call read-only tools against a live system. Hyperglance ships an MCP server that exposes seven FinOps-focused tools. Your assistant decides which to call based on what you ask.

Once connected, you can ask things like “what's driving my Azure bill this month” or “do we have any breached budgets” directly in Claude Desktop, and it will query your actual Hyperglance instance for the answer.

Setting up Hyperglance MCP

The Hyperglance MCP capability can be run in two different ways depending on where you would like to host it or which protocol you would prefer to use:

1) Streamable HTTP protocol:  The Hyperglance MCP service can be enabled in your existing Hyperglance deployment and your AI client will communicate with it over Streamable HTTP.

2) STDIO protocol: The Hyperglance MCP server can be run locally as a docker container on your PC or Mac and your AI agent can communicate it with it over STDIO.

This guide covers both options and provides the typical mcpServers JSON configuration block required but specific steps vary between AI clients, please consult the documentation for your preferred client to learn where the JSON block (or the individual values within it) need to be applied for your preferred agent or chatbot.

Enable the MCP service for Streamable HTTP

1. SSH into your Hyperglance deployment

2. Edit the /var/lib/data/hyperglance_helm_charts/values.yaml config file

3. Enable the mcp server by setting the ai.mcp.enabled flag to true:

ai:
   mcp:
     enabled: true

4. Create an API Key for Hyperglance

5. Connect any MCP-compatible client that supports the Streamable HTTP protocol by using this configuration:

• Replace YOUR-API-KEY with your actual API key

• Replace your-hyperglance.company.com with the IP or Hostname of your Hyperglance deployment.

{
  "mcpServers": {
    "hyperglance": {
      "url": "https://your-hyperglance.company.com/mcp",
      "headers": {
        "Authorization": "Basic YOUR-API-KEY"
      }
    }
  }
}

If your MCP client doesn't support custom headers or if you would rather move the secret into the Helm configuration or into a Kubernetes Secret you can instead configure the ai.mcp.hgApi fields in the /var/lib/data/hyperglance_helm_charts/values.yaml file.

Run the MCP docker container locally for STDIO

1. Run Docker Desktop on your PC or Mac

2. In the MCP configuration of your preferred AI client use this configuration:

• Replace YOUR-API-KEY with your actual API key

• Replace your-hyperglance.company.com with the IP or Hostname of your Hyperglance deployment.

{
  "mcpServers": {
    "hyperglance": {
      "command": "docker",
      "args": [
        "run", "-i", "--rm",
        "-e", "HG_URL=https://your-hyperglance.company.com",
        "-e", "HG_AUTHORIZATION=Basic YOUR-API-KEY",
        "hyperglance/mcp"
      ]
    }
  }
}

 

What tools does the MCP server expose?

The MCP server exposes several tools.  For a fully up-to-date list just ask your AI agent or chatbot for the available tools.

What sort of questions can I ask?

Here are some example prompts to get your started:

• List all EC2 instances tagged env:prod
• What's driving my AWS bill this month?
• Are any of my budgets at risk?
• What compliance violations are currently open?
• Which clouds do we have connected?
• What resource types can I query for Azure?
• Where can I save money on my GCP spend?

 

Security

• API keys grant full HGAPI access. The MCP tools themselves only read data, but the API key backing them could do more in other contexts.  Treat that API key like a password.
• Only a Hyperglance administrator can create or delete API keys, from Settings → API Keys. Deleting a key revokes it immediately for anything using it, MCP included.
• Only give this key to assistants and users you'd trust with the underlying data: query results (resource details, costs, findings) are sent to whichever AI assistant you've connected.
• Everything runs inside your own self-hosted instance. No inventory or cost data is sent to Hyperglance or any third party as part of enabling this feature but the data is of course shared with your AI agent.
• API keys grant full HGAPI access. The MCP tools themselves only read data, but the API key backing them could do more in other contexts.  Treat that API key like a password.

 

Troubleshooting

401 Unauthorized	The Authorization value is missing, mistyped, or the key was revoked. Re-copy it from Settings → API Keys, and paste it whole, including Basic .
404 on /mcp	The MCP server isn't enabled on this instance yet. Ask your Hyperglance administrator to set ai.mcp.enabled: true.
No tools appear in Claude Desktop	The config file has a typo, or Claude Desktop wasn't fully restarted after editing it. Quit the app completely and reopen.
“No resources found” for a query that should match	Confirm the datasource name with “which clouds do we have connected?” first: names are case-sensitive (e.g. Amazon, not AWS).
Request times out on a very broad question	Large estates can take a while to summarize. Narrow the question (a specific datasource or account), or ask your administrator to raise the server's timeout.