Rootshell Platform · MCP Server · Beta
Bring your own AI to your exposure data.
Connect your own LLM to the Rootshell Platform through our MCP Server. It reads your findings, exploit intelligence and asset context, joins them to what you know about your business, and shows what the records might mean together. People still decide what happens next.
Start with the platform
The platform tells you which findings are dangerous.
The Rootshell Platform already does a great deal of the work needed to turn security data into action. It brings scanner results, penetration tests and assessments into one place, tracks recurring findings across scans, and prioritises issues using severity, asset priority and your own tags. You manage remediation through workflows, integrations and live dashboards, including SLA and time-to-fix reporting.
Exploit Detection marks the findings that matter most at a glance:
Black bug
A known working exploit exists for this vulnerability.
Red bug
The vulnerability is being actively exploited in the wild right now.
An LLM needs a reliable starting point. It cannot make up for missing evidence by sounding confident. Its value comes from working across the evidence the platform holds and the extra context you have about your own environment.
The attack path that caught our attention
Why a lower-priority system can still be fix-first.
In our test environment the most interesting output was a possible route from a lower-priority endpoint towards a high-priority cloud production account. The endpoint had a red-bug browser flaw listed in CISA’s Known Exploited Vulnerabilities catalogue. The platform identified the finding; the LLM asked what that endpoint might lead to.
Endpoint browser flaw towards the cloud production account
A view the LLM built for itself from platform records. Test data. Every step is graded by the evidence behind it. Nothing was tested.
Four of the seven steps were not evidenced, and the LLM said so. It labelled the path plausible, not tested. That distinction is the point: it did not discover a working attack, it gave our testers precise questions to settle. Who uses the endpoint? Which browser and controls are present? Which accounts are reached from it, and what privileges and sign-in protection do they have? It did not change the priority rating or reorder the fix list.
Beyond the platform
Join the records to your own context, and to public intelligence.
Your cloud and firewall exports
We gave the LLM an export of cloud servers and firewall rules. It compared them with the platform’s asset records and scan history and highlighted internet-facing servers whose firewall rules allowed remote access or web traffic, despite recent scans finding nothing significant. That is a useful question for a tester: were the scans in scope for those services? It also surfaced a server missing from the asset register, and a server recorded as internal that the cloud export showed with a public address. Once our team confirmed they were the same machine, it recommended moving the finding to the top of the fix list. Your team decides whether a record or a priority changes.
Public sources, cited
The platform’s black and red bugs say an exploit exists or is being used. The LLM then went looking for the wider story, using public sources it could cite. It checked each flagged flaw against CISA’s Known Exploited Vulnerabilities catalogue and read public threat reporting for any group named as using the flaws, putting the source and date beside each name. Two rules keep it honest: every item carries its source so you can check it, and attribution is context only. It never changes a priority, and it never implies you are being targeted unless a source says so.
One analysis, different questions
The same evidence, explained for whoever is asking.
Testers
What should I verify next, and why?
Remediation owners
Which action breaks a possible route?
The board
A short account of exposure and progress.
The CISO
Which threat actors are exploiting the flaws on our systems today, and are they going after our sector? The answer comes from your own open findings and cited sources, and says plainly when no source names your sector.
This does not replace the platform’s live dashboards, workflows or metrics. It makes it easier to interrogate them alongside other information, and to turn a specific question into a focused worklist.
| The Rootshell Platform provides | An LLM connected through the MCP Server can add |
|---|---|
| Consolidated findings and daily exploit intelligence | Questions about how findings may relate across assets and accounts |
| Client-defined asset priorities, owners and tags | Checks where other evidence suggests context should be reviewed |
| Finding history, remediation workflows and dashboards | Explanations and worklists tailored to the person asking |
| A continuing record of what has been found and fixed | Comparisons with cloud configuration or other data you supply |
The boundary matters
An LLM can suggest where to look. A person decides whether to touch a live system.
It is now technically possible to give an LLM tools that scan, probe and try to exploit live systems on its own. Our answer is no. For live client assets, human-led is the safest approach, and it is the one our clients choose. When our testers find something that could be taken further, they stop and ask: do you want us to go further? That decision belongs to you and to an accountable tester, not to a model acting on its own.
LLMs are not always right, but they can sound convincing when they are wrong. In our test the LLM sometimes concluded that finding history or SLA measures were missing, when those capabilities already existed in the platform. A fluent answer is not evidence. The MCP Server is a read-only window, and the LLM sees only what it returns, which may be less than the platform itself knows. If you connect an LLM to your own data, do not assume it is reading everything.
The LLM reads, correlates and proposes
Paths, gaps, questions and recommended priority changes, each with its evidence and its unknowns.
People check
Conclusions are verified against the underlying records before anyone acts on them.
People decide
You and your testers decide whether a record or a priority should change.
People test
Where a question can only be settled by touching a live system, the LLM stops and hands it to an authorised tester, working within your permission.
Human-led. AI-augmented. Continuous.
See what an LLM could find in your own data.
The Rootshell Platform MCP Server is in beta and coming to clients soon. As we develop it, we are exposing more finding history and asset relationships, so you spend less time working out where to look and more time reducing the exposure that matters.
Explore the platformMethod: data analysis of a test environment only. No live checks, probing or exploitation were performed. All systems, figures and counts are test data. Public vulnerability identifiers, CISA KEV dates and threat-actor attribution are taken from their public sources (CISA, NVD and Volexity).