Trust Center: How ClevSec Handles Your Data and Secures Its Tools

Two places our tools run

	Website tools	MCP server for AI assistants
Where it runs	In your browser (static pages on Cloudflare Pages)	On a Cloudflare Worker at mcp.clevsec.one
Your answers	Never leave your device. Exports (PDF, Excel) are generated locally.	Your assistant sends the facts a tool needs. We compute the answer and do not store the request.
What we log	Nothing about your answers. Cookieless, aggregate Cloudflare Web Analytics for page views.	Which endpoint and which tool was called. Never the tool arguments.
AI inside	None. Deterministic rules.	None. The same rules, from the same code as the website tools.
ClevSec Pro	Your browser sends the licence key straight to LemonSqueezy. We only keep a "Pro active" flag in session storage, never the key.	OAuth sign-in with your licence key, stored encrypted in your access grant (details below).

The tools ask for facts about an organisation or an incident (a sector, a headcount, how long an outage lasted), not for personal data. Please do not include any.

MCP server security

Read-only tools. Every tool is declared read-only, non-destructive and idempotent. A tool can compute an answer; it cannot change anything. The only outside call is the Pro licence check with LemonSqueezy.
No language model on the server. Answers come from fixed rules that are tested before every deploy, so a prompt cannot talk the server into a different result.
Validated input. Each tool has a JSON Schema. Invalid input gets a precise error back; any other failure returns a generic message without internal details.
Stateless free endpoint. /mcp needs no account and keeps no session data.
Abuse limits. 300 calls per minute per IP address. Cloudflare processes the connecting IP address to enforce this.
Outbound requests limited to public addresses. The Worker can only fetch public internet addresses, which protects the sign-in flow against requests to internal networks.

ClevSec Pro sign-in for AI assistants

The Pro address (/pro) uses OAuth 2.1, as the MCP authorization specification requires.

You enter your licence key once, on our sign-in page. We check it with LemonSqueezy before granting access.
The key is stored only inside your access grant, encrypted so that only your assistant's token can unlock it. Tokens are stored hashed.
Access tokens last one hour; refresh tokens last 30 days and expire after 30 days without use. The grant is then deleted.
We re-check your subscription with LemonSqueezy at most every ten minutes, so a cancelled subscription stops working within minutes. That cache is in memory and keyed by a hash of the licence key.
The sign-in page shows where access will be sent, warns about unverified apps and apps on your own computer, and cannot be framed by other sites. ClevSec never sends sign-in links by email.
We do not ask for or store your name or email for this.

Website security

Static pages only, so there is no server-side code that handles your answers.
A Content Security Policy that allows only our own scripts and Cloudflare's analytics beacon, and lets pages connect only to our own site, LemonSqueezy (licence checks) and FormSubmit (newsletter form).
HSTS with preload, X-Frame-Options: DENY, nosniff, and a Permissions-Policy that turns off camera, microphone and location.
No cookies, no advertising trackers. Fonts are served from our own site, so your browser does not connect to Google Fonts.

Third parties

Provider	What for
Cloudflare	Hosting, DNS, cookieless analytics, running the MCP server, rate limiting, storing encrypted Pro access grants
LemonSqueezy	Merchant of Record: payments, product delivery, licence key checks
FormSubmit	Newsletter sign-up form

Full details, legal bases and retention periods are in our privacy policy.

Software supply chain

The MCP server has three runtime dependencies (the official MCP SDK, Cloudflare's OAuth provider library and zod). All are pinned to exact versions with a lockfile. Install scripts are allowed only for two named build tools.
Every deploy runs the rule tests and the protocol tests first. The deployed server is then tested again, and it is re-tested every week.
The tool rules are open data under CC BY 4.0, so anyone can check them: clevsec.one/data.

Versions and changes

MCP server version 1.4.0, Streamable HTTP, MCP specification 2026-07-28. Listed in the official MCP Registry as one.clevsec/eu-compliance.
Every change to the rules and tools is recorded in the public changelog.
Live server information: mcp.clevsec.one. Setup instructions: Use ClevSec from your AI assistant.

Report a vulnerability

Email [email protected] with a description and steps to reproduce. Our contact details are also in security.txt (RFC 9116) on both clevsec.one and mcp.clevsec.one. We welcome reports and will answer them, but we do not run a paid bug bounty programme. Please test only against your own account and do not degrade the service for others.

Open client registration on the MCP server is intentional: the MCP authorization specification lets AI assistants register themselves. It is not a vulnerability.

What we do not claim

ClevSec is a small independent studio. We have no SOC 2 report or ISO 27001 certificate, and no third-party audit of these services. This page describes how the services are built. It is not an audit result.