Make every agent you run security-aware
Your agents can talk to Mondoo. You hold the keys to what they do.
Keep the AI agents and assistants your teams already use. Mondoo gives every one of them a trusted source of what's exposed in your environment and how to fix it, through a hosted MCP server that's included for every customer. And it secures the MCP servers your teams connect, too.
- What's exposed
- 3 hosts run OpenSSH 9.6, affected by
CVE-2024-6387 - What matters most
prod-api-02first: public exploit, holds customer data- How to fix it
- Upgrade
openssh-serverto 9.8p1, then rescan
- Works with the agents you already use
- Read-only by default: you decide what agents can do
- Hosted in the US and EU, included for every customer
Mondoo works on both sides of MCP.
Choose a path
Your agents work fast. They don't know what's exposed.
Your teams' AI agents and assistants already write code, open tickets and change configurations. But they don't know what's exposed in your environment, which risks matter or how to fix them safely. So they guess, or they stop and hand the question back to a person.
Mondoo gives every agent the answer, and the fix.
Connect your agents to Mondoo
Don't replace your agents. Give them a source they can trust.
Which agents connect
AI assistants
such as Claude, for security leaders and analysts who want answers in plain language.
AI coding agents
such as Claude Code and GitHub Copilot in VS Code, for developers fixing code where they work.
Your own agents and workflows
through any client that supports MCP.
Mondoo doesn't ask you to switch to its agents. It opens its AI-native platform through MCP, so every agentic tool your teams already use can query Mondoo and act on its findings and fixes.
A trusted source of what's exposed.
Not a raw list of findings. Mondoo has already checked each exposure across code, cloud, endpoints, AI and your external attack surface, and ranked it by Mondoo Risk Dimensions: attack surface, blast radius, exploitability and business priority.
And how to fix it.
Every finding comes with remediation guidance your agents can use: the fix steps, generated code and script snippets, and the policy as code that keeps it fixed.
With guardrails you set.
Agents act only within the permissions of their token. The fix itself runs through Mondoo's agentic remediation, autonomously where it's safe and with your approval where it matters.
One platform's view of what's exposed, and how to fix it, in every agent you run.
See agentic remediationYou hold the keys. Agents only do what you allow.
An agent connected to Mondoo can never do more than its API token allows, and you issue the token. Start read-only. Widen it only where you want an agent to trigger fixes or open tickets, exceptions and workflows. And whatever an agent asks for, the fix still follows your remediation policy, so nothing reaches the systems you've protected without a person's approval. Rotate or revoke a token at any time, and the agent's access ends.
Read-only by default.
A Viewer token can explore and explain, but can't change anything.
Permissions you set.
Write access comes from the token you issue, never from the agent.
Approval where it matters.
Fixes an agent triggers run autonomously only where your policy allows. Production servers, databases and any system you choose wait for a person to approve the plan, the impact and the rollback.
Every fix on record.
What changed, who approved it and when it was verified is recorded, so you can see exactly what happened and why.
We don't hand AI the keys. We hand you the control.
Ask in plain language. Get answers from your own data.
The Model Context Protocol (MCP) is the open standard AI assistants use to work with other tools. Through it, your assistant works directly with Mondoo Platform.
Find what's exposed.
List critical findings in a space, or search across assets, policies and findings.
“Which internet-facing assets have critical findings?”
Understand why it matters.
Get an explanation of what a policy checks and why a finding matters for your environment.
“Why is this finding critical, and what does the policy check?”
Stay ahead of what's trending.
See the CVEs trending in the news and on social media, and whether they affect you.
“Are we exposed to any of this week's trending CVEs?”
Start the fix.
With the right permissions, trigger the fix, or create tickets, exceptions and workflows for the findings your assistant surfaces.
“Fix the critical finding on prod-web-01.”
Write policy as code.
Generate MQL queries and policies from a description, then compile and validate them.
“Write a policy that checks SSH is hardened on our Linux servers.”
You
What are the critical findings in our production space?
Mondoo: search findings
3 critical findings across 12 assets. The highest risk is an internet-facing server with a known exploited vulnerability.
- Known exploited vulnerability in web server packageprod-web-01Critical
- Remote code execution in Java logging libraryprod-api-03Critical
- Privilege escalation in Linux kernelprod-worker-07Critical
You
Fix the first one.
Mondoo: trigger fix
Done. A fix is planned for prod-web-01. It's a production server, so it's waiting for your approval before it deploys.
An MCP server you don't have to run yourself
Many security MCP servers are something you install, run and keep updated on your own machines, or a preview you request access to. Mondoo's is a hosted service, ready for every customer, and connected to the fix.
Hosted for you, in the US and the EU.
No local server to install, patch or keep running. Connect over HTTP to the endpoint for your region, including an EU endpoint for EU accounts.
Ready now, for every customer.
Generally available and included with Mondoo at no extra cost. Not a preview, and not an add-on.
Read-only by default.
Start with a Viewer token that can explore and explain but can't change anything. Grant more only where you choose.
Connected to the fix.
Most MCP servers stop at answering questions. With Mondoo, your assistant can trigger the fix, and Mondoo's agents carry it through, autonomously where it's safe and with your approval where it matters.
How Mondoo's hosted MCP server compares with the security MCP servers other tools ask you to install and run yourself.
Mondoo MCP serverSet up in minutes
The Mondoo MCP server works with Claude Desktop, Claude Code and VS Code, and with other assistants that support MCP.
Create a token.
Create a read-only Viewer token in Mondoo.
Add the server.
Add the Mondoo MCP endpoint for your region to your assistant.
Ask.
Ask your assistant to list your Mondoo spaces. If it does, you're connected.
Secure the MCP servers you use
MCP server security: protect every server your agents connect to
Every MCP server your teams connect to an AI assistant is a new way in. It can expose tools that change systems, prompts that can be manipulated and data that shouldn't leave.
The risks hiding in an MCP server
- Prompt injection and tool poisoning.Hidden instructions in a tool's description or content tell the agent to do something the user never asked for.
- Over-permissive tools.Tools with no limits on their inputs can be pushed to do far more than intended.
- Sensitive data exposure.Personal data or secrets in a server's content or resources reach every agent that connects.
- Hidden characters and lookalike text.Invisible or lookalike Unicode characters disguise instructions and names that a person reviewing the server would miss.
- Unvetted links.URLs embedded in tool content can send an agent, or the person using it, somewhere they shouldn't go.
- Shadow servers.MCP servers nobody approved, running on laptops and reaching production with the user's credentials.
Find them.
Mondoo discovers the MCP servers, AI agents and skills running across your fleet, including the ones nobody approved.
Assess them.
cnspec scans local and remote MCP servers against the Mondoo MCP Security policy, which reviews the tools, prompts and resources each server exposes.
Enforce what's allowed.
Mondoo enforces what may run: banned agents don't operate and malicious skills never run. Every MCP server is monitored continuously, and findings close through the endpoint tools your teams already use.
What the Mondoo MCP Security policy checks
Seven checks, built in and maintained by Mondoo, each based on established guidance such as OWASP.
Need more? Write your own checks in MQL and add them alongside Mondoo's.
| Check | What it catches |
|---|---|
| Prompt injection detection | Instructions hidden in tool descriptions or content that try to take over the agent |
| Tool parameter validation | Tools whose inputs have no validation constraints, so they can be pushed beyond their purpose |
| PII detection | Personal data in tool definitions, prompts or resources |
| Resource and template validation | Resources and resource templates that expose more than they should |
| Unicode validation | Hidden or lookalike characters used to disguise instructions or names |
| Embedded URL detection | Links in MCP content that could redirect an agent or a user |
| Content filtering | Inappropriate or explicit content in tool definitions |
- Tool has no description
- Tool input declares no required parameters
Scan an MCP server in one command
Free with open source cnspec on your own machine, or run the same checks continuously across your fleet in Mondoo Platform.
cnspec scan mcp http https://mcp.example.com/mcp
Frequently asked questions
Using the Mondoo MCP server
MCP server security
Keep your agents. Make them security-aware.
See what your agents could ask Mondoo, and what Mondoo could fix.
Already a customer? Read the setup guide
Fix Now. Fix Forever.