Found a security issue? We want to hear from you.
Security research makes FindAgent safer for everyone who connects an agent. If you believe you’ve found a vulnerability, report it to us privately and we’ll work with you to fix it. This page explains how to reach us, our good-faith commitment to researchers, and what’s in scope.
How to report
Email us at security@findagent.cloud with enough detail for us to reproduce the issue: what you found, the steps to trigger it, and its impact. Please report privately and give us a reasonable chance to fix the problem before sharing it publicly. Don’t access, modify, or delete data that isn’t yours, and don’t run tests that could degrade the service for other users.
Our machine-readable contact details are published at /.well-known/security.txt.
Safe harbor
We will not pursue or support legal action against you for security research conducted in good faith under this policy. That means: you make a genuine effort to avoid harm to our users and their data, you stay within the scope below, you don’t exfiltrate more data than is needed to demonstrate an issue, and you give us a reasonable time to respond before any public disclosure. If you’re acting in good faith, we consider your research authorized and we’ll work with you. If you’re unsure whether something is allowed, ask us first at security@findagent.cloud.
Scope
In scope
- The FindAgent web application and its APIs.
- The hosted MCP gateway that serves agents.
- Authentication, session handling, and access control.
- The agent-review and publish pipeline, and the sandbox that runs hosted code agents.
- How credentials you give an agent are stored and where they are sent.
Out of scope
- Third-party services an agent connects to — report those to the provider that runs them.
- External MCP servers listed only as a pointer; FindAgent does not run or inspect them.
- Reports produced solely by automated scanners with no demonstrated impact.
- Denial-of-service, volumetric, or social-engineering attacks against our staff or users.
- Missing security headers or best-practice suggestions without a concrete, exploitable impact.
What to expect from us
We aim to acknowledge a report within a few business days, keep you updated as we investigate, and let you know when the issue is resolved. We’ll credit you for a valid report if you’d like the recognition. FindAgent is a small team and we don’t currently run a paid bounty program, but we genuinely value the help and treat every report seriously.
What we’ve found and fixed
A plain-language, sanitized history of security improvements we’ve shipped. We leave out any detail that would help someone attack an older version.
- Info
Made every failure diagnosable the instant it happens
We swept the platform so that when something goes wrong on a serve or connect path, our monitoring already records exactly what failed and why. This means we detect and respond to a problem the moment it occurs, instead of waiting for someone to report a silent outage.
- Low
Personal data can now be stripped from an agent’s results
Alongside the always-on scan that blocks leaked secrets in a result, an agent can now enable an extra rule that removes personal data from what it returns to you. The baseline secret-and-injection scan still runs first and can never be switched off.
- High
Tightened credential-to-host binding for every connection type
A credential you give an agent — an API key or token — is bound to the exact hosts that agent declares up front, and is only ever attached to a request going to one of those hosts. We extended this so it holds across every way an agent runs, including self-hosted targets you point it at, so a secret can never reach a host you didn’t consent to.
- Medium
Closed a gap that let a manifest declare a tool twice
Our checks now reject an agent definition that lists the same tool name more than once. This removes any ambiguity about which safety rules apply to a tool, so the strictest declared protection always wins.
- High
Hosted code agents run with the network locked shut by default
When a hosted code agent runs, it starts in a single-use sandbox with no network access at all and can only reach the specific hosts the agent declares. The sandbox is created fresh for each run and thrown away afterwards, walled off from FindAgent’s own systems.
- Medium
Enforced a strict content-security policy across the site
We turned on a browser-level policy that blocks scripts and connections from sources we haven’t explicitly allowed, and reports any attempt to violate it. This hardens every page against a class of code-injection attacks.
- High
Made the publish gate fail closed
Before any agent goes live it must pass an automated security scan and a human review, and that decision is enforced on our servers. Anything the checks can’t verify as safe is now held back rather than waved through — an agent can never publish itself past a failed scan or a failed build.
Want the full security model?
See exactly what every agent can and can’t do, and where your keys go.