AI Penetration Testing with YasnaMCP: from reconnaissance to security report, from your own IP

AI-driven penetration testing: ask in plain language, receive a professional report

A penetration test usually means days of a specialist’s time, dozens of command-line tools and a report that arrives late. With YasnaMCP you tell the assistant “run a penetration test on example.com and report the open ports and vulnerabilities”. It runs well-known security tools step by step on your connector, explains the findings and finishes with a security report ready to hand over.

Penetration testing is only permitted on assets you own or have written permission to test. Scanning or attacking someone else’s systems without authorisation is a crime. YasnaMCP is built for security teams, network administrators and security service providers.

Why from your own IP?

Scans do not run on YasnaMCP’s servers; they run on your connector — the software you installed on your organisation’s Windows or Linux machine. That means:

• Test traffic leaves from your own, known IP and fits your firewall rules and testing agreements.
• Internal networks (for example 10.0.0.0/8) can be tested too, without opening any port to the outside.
• For every request you choose which connector (which machine, which IP) tests which target. Your organisation is always identified from the access token, never from the request body.

Three capability tiers, each its own permission

Each tier is a separate scope, so an organisation gets exactly the level it is authorised for. The tiers escalate:

Engines and tools: industry standards, not reinvention

YasnaMCP did not build a penetration-testing engine from scratch; it orchestrates the best existing ones:

• Core tools: nmap, nuclei, subfinder, httpx and Metasploit.
• Strix: an open-source multi-agent framework for intelligent penetration testing that automates deep assessment and exploit validation.
• HexStrike AI: an MCP server with more than 150 security tools, reached by the assistant through the connector’s local MCP bridge.
• Security skills: penetration-testing knowledge packs that teach the assistant the right method and safe defaults for each tier.

Ask the assistant once to run pentest_bootstrap: it reports which tools are installed on the connector and installs the engine where it can.

Automated engagement: recon to assessment in one request

Say “run a full engagement on 10.0.0.5” and YasnaMCP runs every reconnaissance stage and then every assessment stage on the connector, in order, returning each stage’s result separately in one consolidated answer. The important safety point: an engagement never enters the exploit stage, so an assessment cannot escalate itself into an attack.

Controlling the exploit tier: several locks in a row

• The exploit permission is never offered to any assistant by default; it must be granted deliberately.
• For the in-portal assistant there is a separate per-organisation exploit switch that only platform staff can change, in tenant management; an organisation cannot turn it on for itself. Every change is audited.
• Exploits run only against a single, named, confirmed target.
• Every input — the target, module options and post-exploitation commands — passes the same guard, so no extra command can be injected into the Metasploit console.
• Every penetration-testing call is audited, and rate limits apply both on the connector and at the gateway.

Security report and responsible disclosure

A penetration test ends with a report. Ask “build a security report from the findings” and you get a structured report with the target, the scope and the findings grouped by severity, citing CVEs and exploit sources beside each finding — ready for a client or for management.

If you found a vulnerability in a third party’s system, the assistant can prepare a confidential coordinated-disclosure notice, so the owner can fix the problem before anything is made public.

The portal also lists a curated set of trusted vulnerability and exploit sources, including Exploit-DB, NVD, CVE.org, GitHub Security Advisories, Rapid7, OSV and the CISA catalogue of known exploited vulnerabilities.

Pricing: transparent, per operation

Each penetration-testing operation has a fixed price drawn from your organisation’s credit wallet: the amount is reserved before it runs, settled afterwards, and released if the operation never reached the connector. Security reports and disclosure notices are free. For organisational pricing and security-service packages, contact us.

Where do I start?

Example requests to the assistant:

• “Run a penetration test on example.com and report the open ports and vulnerabilities.”
• “Run a full engagement on 10.0.0.5.”
• “Build a security report from the findings.”

Frequently asked questions

Could the assistant attack a system without my permission?

No. The exploit tier is off by default, is never offered to the assistant, and an automated engagement never enters it. Even when it is on, it runs against one named target only.

Will the scans take my service down?

Reconnaissance is strictly read-only. Assessment is rate-limited to keep load on the target low; even so, schedule tests of sensitive services for quiet hours.

Which assistants does it work with?

The assistant inside the YasnaMCP portal and any MCP-compatible client your organisation has authorised.

Can internal networks be tested?

Yes. Because scans run on a connector inside your network, internal addresses can be tested without opening any port to the outside.