Is it safe to use third-party tools with Claude Code?
It depends on what the tool actually does with your code. The question that matters is whether it executes your code on its own infrastructure or only coordinates around a session running on your machine. Ask where execution happens, what credentials it holds, and how tenant data is isolated.
The distinction that decides it.
Tools in this category fall into two groups, and they carry completely different risk profiles regardless of how similar the marketing sounds.
Tools that execute your code
A hosted agent needs your repository in its environment to work. That is a legitimate model, but it means your code is on somebody else infrastructure and your trust question is about their runtime, their isolation, and their retention.
Tools that coordinate around your session
A control plane observes and relays. Your code never moves, your credentials never move, and execution stays on your machine. The trust question narrows to what coordination data they hold and how it is isolated.
Both can be safe. They are not the same question
The mistake is evaluating them with one checklist. Ask which model you are looking at first, then ask the questions that model actually raises.
Four questions worth asking.
Of us, and of anything else you are evaluating.
Where does execution happen?
On your hardware or theirs. This single answer determines most of the rest of your risk assessment.
Who holds the credentials?
Whether your model subscription and machine credentials stay with you, or are proxied through a vendor.
How is tenant data isolated?
Application-layer filtering fails when a query is written wrong. Ask whether isolation is enforced in storage.
What is recorded, and can you prove it?
Audit logging with actor, target, and timestamp is what makes an agent action attributable after the fact.
How Virex answers each one.
Stated plainly so you can compare it against anything else on your shortlist.
Related questions
What should I ask before connecting a tool to Claude Code?
Three things: does it execute my code or only coordinate, does it hold my model credentials, and how is my data isolated from other customers. Vague answers to any of those are the answer.
Does Virex execute my code?
No. Claude Code keeps running on your machine under your own model subscription. Virex carries session activity, decisions, and memory. There is no Virex environment that builds or runs your repository.
What does Virex actually store?
Session activity and decision records, the memories your agents write, and account and entitlement state. Not your repository, not your credentials.
Is an MCP server a security risk by itself?
MCP is a calling convention and says nothing about access control on its own. What matters is what the server behind it does. With Virex, every tool call resolves inside the calling member organization only.
How do I verify any of this?
Read the security page, which states the mechanisms rather than adjectives, and ask us directly about anything it does not cover. We would rather answer a hard question than lose you to a vague one.
Ask us the hard version.
If your review needs something this page does not cover, ask directly and we will answer specifically.