Agent Verify

Agent Verify 2.0

What's new in 2.0

Agent Verify now checks AI agents everywhere they are built and run — in your editor, on every pull request, and in production — with 124 tested checks.

Catch more

Deeper coverage for the way agents are built today.

  • 48 new checks — 124 in totalNew

    MCP tool poisoning, hidden characters and tool shadowing; payments and deletions with no limit or approval; personal data sent to models; retrieval with no per-customer filter; AI agents in CI; and checks for 11 frameworks including LangGraph, LlamaIndex, the Vercel AI SDK, Semantic Kernel, smolagents and Google ADK.

    Browse the check library →
  • A help page for every checkNew

    What each check finds, why it matters, what an attacker could do, and how to fix it — linked from every finding in your reports.

    Open the check library →
  • State of MCP Security reportNew

    What 520 popular MCP servers can actually do on your computer, and how to use them safely.

    Read the report →
  • MCP security directory

    "Is it safe?" pages for 520 MCP servers, with category comparisons and a pinned install snippet.

    Find a server →

Check every change

Security checks where code is written and reviewed.

  • Pull request checks

    GitHub App, GitLab and Bitbucket: a pass/fail check and one updating comment on every PR that touches agent code. Pro adds one-click fixes.

    Set up PR checks →
  • VS Code and your AI assistant

    Risky lines underlined as you type, and an MCP server so Claude Code, Cursor and Copilot can scan your agents for you.

    Install in your editor →
  • CLI and GitHub Action v2

    Run the same checks locally and in CI, with each finding listed in the PR comment.

    Use the CLI →

Protect what runs

Controls for agents in production.

  • Agent Guard

    Runtime rules for JavaScript/TypeScript agents: allow and deny lists, human approval for risky tools, allowed domains, rate limits, and blocking keys from leaving.

    Add Agent Guard →
  • Prompt-injection tests (Pro)

    Ten known attacks sent to your own agent, with clear results on which ones worked.

    Learn more →
  • Compliance evidence pack (Pro)

    EU AI Act and ISO/IEC 42001 mappings plus a one-click .zip of evidence for auditors.

    See compliance →

Easier for teams

Simpler to start, share and run across a team.

  • Team plan

    Per-seat checkout, every Pro feature, and teammates managed from Billing.

    See plans →
  • Ask anything (Ctrl+K)

    One box on every page to find answers, docs, checks, MCP servers and your own findings.

    Try it: press Ctrl+K →
  • Slack, alerts and a weekly email

    Scan from Slack, get alerts only for new risks, and a short Monday summary.

    Connect Slack →
  • New look, Trust Center and Status

    A cleaner interface, plus clear answers on what we store and live system status.

    Visit the Trust Center →

Every change, with details: full changelog on GitHub.