Product · CloudFix MCP Server

CloudFix MCP Server: AWS Cost Optimization for the AI Age

No dashboard required. Connect Claude, Cursor or any MCP client to CloudFix and ask what is wasting money in your AWS accounts, in plain English. When you are ready to act, CloudFix makes the change, once you approve it. Free on every CloudFix plan.

6 read-only tools3 ready-made client configs$0 on every plan

Claude Code · cloudfix-recommendations

What are our biggest AWS savings right now, and which can CloudFix fix on its own?

get_recommendations_summary

list_recommendations · sort_by: annualSavings

Three optimization types make up most of it:

  1. EC2 instance optimization, 14 instances $38,400/yr
  2. EBS gp2 to gp3, 212 volumes $21,900/yr
  3. Idle load balancers, 9 $6,300/yr

The gp2 to gp3 and idle load balancer fixes support auto-fix: approve them in CloudFix and they run in your account. The EC2 changes are best done in a maintenance window. Want the analysis report for those?

Illustrative example. Figures are not from a customer account.

Why now

If it needs a dashboard, AI-native teams won’t open it.

The people building with AI every day have said it plainly: software should come to the assistant, not the other way round. Cost tools are no exception. The answer to “what is wasting money?” belongs where the work already happens: next to the code, in the chat where the update gets written, inside the agents teams build for themselves.

  • “Don’t talk to a developer. Don’t ask a developer to visit, look, or click. Instruct and empower their LLM.”

  • “Why should you ever log into Salesforce again?”

    Parker Harris, Salesforce co-founder · April 2026
  • “The agent never opens your app.”

    Navin Chaddha, Managing Partner, Mayfield · May 2026

How it works

Ask in your AI. Fix in CloudFix.

Your assistant does the part assistants are good at: finding, explaining, ranking and writing it up. CloudFix does the part that touches your account, the way it always has: tested fixers, your approval, AWS’s own tooling in your account.

  1. 01Ask

    Your assistant reads CloudFix.

    Ask in plain English. Your assistant calls CloudFix’s MCP tools to pull the recommendations, savings and reports CloudFix has already found in your AWS accounts.

    • No dashboard, no query language
    • Filter by account, region, organizational unit, status or finder
    • The same recommendations you see in the CloudFix app
  2. 02Decide

    It explains, ranks and drafts.

    Your assistant puts CloudFix’s findings next to the context it already has: your code, your tickets, your notes. Ask it to rank by savings, explain a finding, or write the update for leadership.

    • Sort by annual savings or annual cost
    • The full analysis report behind any recommendation
    • Which savings CloudFix can fix automatically
  3. 03Fix

    CloudFix makes the change.

    Fixes happen in CloudFix, never from the chat window. Once you approve, each fix runs as an AWS Systems Manager Automation runbook in your own account, and every execution is logged.

    • You approve every change
    • Run now or schedule a day and time
    • 53 automated fixers, with a snapshot first where the fix supports it

What CloudFix won’t automate, and why

The tools

Six tools. Every one of them reads.

Your assistant chooses the tool; you just ask the question. Each tool can be switched on or off for your account in MCP Settings.

  • list_recommendations

    Every savings opportunity CloudFix has found. Filter by finder, status, AWS account, region, organizational unit, date or resource, and sort by annual savings or annual cost.

    Ask “List my ten biggest recommendations by annual savings in us-east-1.”

  • get_recommendation_details

    Everything about one recommendation: the resource, its parameters, its status and the savings attached to it.

    Ask “What exactly would change on this volume, and what does it save?”

  • get_recommendation_report

    The finder’s detailed analysis report behind a recommendation: the evidence it acted on.

    Ask “Show me the evidence behind this rightsizing recommendation.”

  • get_recommendations_summary

    Counts and potential, completed and realized savings, grouped by optimization type, with whether each type can be fixed automatically.

    Ask “Summarize our AWS savings for this week’s leadership update.”

  • get_ec2_price

    The hourly, monthly and yearly price of an EC2 instance type, on demand or on a 1- or 3-year Savings Plan term.

    Ask “What does an m7g.2xlarge cost in eu-west-1 on a 1-year plan?”

  • find_ec2_instances

    EC2 instance types that meet a spec (vCPU, memory, GPU, processor, family or a price ceiling), ranked by price.

    Ask “Cheapest Graviton instance with 8 vCPUs and 32 GB in us-east-1?”

More questions to ask your AWS bill

Who it is for

Cost answers where the work already happens.

The CloudFix app is still there when you want it. The MCP server means you no longer have to go to it to know where your AWS bill stands.

  • Engineers in Claude Code or Cursor

    Price an instance type before it goes into Terraform, and see what CloudFix has flagged in the account you are working in, without leaving the editor.

  • FinOps and platform leads

    Run the weekly review as a conversation: what is new, what is waiting in Manual Approval, what failed and what has already been fixed.

  • CTOs and finance

    Get the savings picture in a sentence, or a slide’s worth, in the assistant you already use. No login, no export, no screenshot of a chart.

  • The agents you build

    Any MCP client can connect, so your own agents and workflows can read CloudFix recommendations the same way your assistant does.

Set up

Connected in three steps.

  1. Check it is on

    In the CloudFix app, open MCP Settings and check that Enable MCP Integration is on (it is by default). Choose which tools to allow while you are there.

  2. Generate a key

    Click Generate MCP Key. CloudFix mints a Reader API token that starts with cft_. It is shown once, so copy it before you close the dialog.

  3. Add the server to your client

    For Claude Code, run the command below. MCP Settings has the same snippets for Cursor and Claude Desktop, with your key already filled in. Then run /mcp in Claude Code to check the connection.

Then ask: “List my CloudFix recommendations sorted by annual savings.”

Claude Code: one command

claude mcp add --transport http cloudfix-recommendations \
  https://mcp.app.cloudfix.com/mcp \
  --header "Authorization: Bearer <paste-your-cft-token-here>"
Claude Code: config file

Merge into ~/.claude.json, or a project .mcp.json.

{
  "mcpServers": {
    "cloudfix-recommendations": {
      "type": "http",
      "url": "https://mcp.app.cloudfix.com/mcp",
      "headers": {
        "Authorization": "Bearer <paste-your-cft-token-here>"
      }
    }
  }
}
Cursor

Merge into ~/.cursor/mcp.json, or a project .cursor/mcp.json.

{
  "mcpServers": {
    "cloudfix-recommendations": {
      "url": "https://mcp.app.cloudfix.com/mcp",
      "headers": {
        "Authorization": "Bearer <paste-your-cft-token-here>"
      }
    }
  }
}
Claude Desktop

Merge into claude_desktop_config.json. Claude Desktop reaches the server through the mcp-remote bridge.

{
  "mcpServers": {
    "cloudfix-recommendations": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.app.cloudfix.com/mcp",
        "--header",
        "Authorization: Bearer <paste-your-cft-token-here>"
      ]
    }
  }
}
Installer for Claude Code

macOS or Linux, with the claude CLI on your PATH. Keep API_KEY on the sh side of the pipe.

curl -fsSL https://app.cloudfix.com/install-mcp.sh | API_KEY=cft_<your-api-token> sh

Security model

An assistant can read. Only you can approve.

Giving an AI assistant access to your cost data should not mean giving it the keys to your AWS accounts. With CloudFix, it never does.

  • Read-only by design

    Every CloudFix MCP tool reads. None can approve, run or change anything in your AWS accounts.

  • A Reader key you control

    The key is a Reader API token (it starts with cft_), shown once when you create it. Revoke it at any time in API Tokens.

  • You choose the tools

    Switch MCP off for the whole account, or turn individual tools on and off in MCP Settings.

  • Your tenant, and only yours

    Each request is scoped to the CloudFix tenant that owns the token. The server reads CloudFix’s data, not your AWS resources.

  • Changes still need approval

    Fixes run in CloudFix after someone you allow approves them, as AWS Systems Manager Automation in your own account.

How CloudFix accesses your accounts

FAQ

Frequently asked questions.

Setting it up for a team? Talk to us.

Is the CloudFix MCP server free?

Yes. It is included with every CloudFix plan, including the free tier.

Can my AI assistant change my AWS account?

No. Every CloudFix MCP tool is read-only. Your assistant can find, explain and prioritize savings; the change itself is approved and run in CloudFix, as AWS Systems Manager Automation in your own account.

Which AI assistants work with it?

Any MCP client that can connect to a remote HTTP server, or reach one through a bridge such as mcp-remote. MCP Settings has ready-made snippets for Claude Code, Cursor and Claude Desktop.

What can the assistant see?

Your CloudFix recommendations, their details and analysis reports, savings summaries, and CloudFix’s EC2 pricing data. The server reads from CloudFix; it does not connect to your AWS resources.

Where do the answers go?

To the AI client and model you connect, under that provider’s terms. Turn off any tool you would rather keep out of an assistant’s reach in MCP Settings.

Who controls it?

MCP Integration is on by default. An administrator can switch it off, or change which tools are allowed, in MCP Settings. Creating a key needs permission to create API tokens.

What is MCP?

Model Context Protocol is an open standard that lets AI assistants call external tools and data sources. An MCP server exposes the tools; clients such as Claude and Cursor connect to it.

Give your assistant something worth reading.

Connect your AWS accounts with read-only roles in about 5 minutes. Recommendations typically within 24 hours, ready for any MCP client.