- Spot attempts to redirect your agent away from your request
- Help prevent prompt injection, tool poisoning, and agent hijacking
- Keep sensitive actions under your control
- Help you understand what happened after an action
WebMCPify Guard · Open source
Protection for the person behind every agent action.
Your agent should work for you—not the page it is reading.
WebMCPify Guard is being built to help protect you when an AI agent uses websites. It aims to block unsafe instructions and keep sensitive actions under your control. Try the Guard Inspector today while broader protection is in development.
DEVELOPER PREVIEW The local Inspector works today. The connection, policy, and execution layers are being built in public.
YOUR TASKYour AI agentStarts with the job you asked it to do
↓SAFETY CHECKGuard checks for riskLooks for misleading instructions and unsafe actions
↓YOUR CHOICEYou stay in controlImportant actions need your say
↓WEBSITEWebMCP-enabled siteThe agent uses approved website capabilities
↓RESULTClear outcomeSee what the website returned
PLANNED GUARD FLOW · INSPECTOR AVAILABLE NOWHow Guard fits
Protect the whole path—not only the approval click.
Guard is designed to help your agent work safely while you stay in control of important website actions.
- 01
Connect
Bring a compatible agent to a WebMCP-enabled website.
- 02
Check
Look for instructions or tools that could pull the agent away from your task.
- 03
Choose
Keep important decisions in your hands.
- 04
Act
Let the agent use the website within the access you approve.
- 05
Confirm
See whether the intended action worked.
- 06
Review
Understand what the agent did and what came back.
Protection + connection
A security system with a bridge inside it.
Connect a compatible agent to WebMCP-enabled sites while keeping control of what it can do in your browser.
- Let compatible agents use WebMCP-enabled websites
- Keep website actions within the access you approve
- Give your agent clear results you can review
Your control stays in the loop. Guard protection applies to actions your agent routes through Guard.
What Guard protects against
Security for the whole agent-to-browser path.
Guard is being built to prevent direct and indirect prompt injection, agent hijacking, tool poisoning, and unsafe website actions. Today, the Guard Inspector helps identify risky tool definitions before your agent uses them.
Injected instructions
Spot when page content tries to give your agent a different job.
Agent hijacking
Keep the original request in focus when a site tries to redirect it.
Tool poisoning
Flag tool descriptions or inputs that hide risky behavior.
Sensitive actions
Make high-impact steps clear before they happen.
Data exposure
Help keep personal information within the approved task.
Misleading results
Make outcomes easier to review instead of trusting a success message alone.
Guard Inspector · Developer Preview
Inspect the tools an agent can actually call.
Discover this page's registered WebMCP tools, inspect schemas and safety hints, validate arguments, and run one controlled call. You can also paste a descriptor for an offline audit. Everything stays in this browser tab.
Reading this page
0 registered tools visible to this document.
Status, without fog
Working baseline. Larger boundary ahead.
Try the Inspector now. Broader agent protection and connection features are still in development.
Developer Preview
- Local tool-definition inspection
- Annotation checks
- Sensitive-parameter warnings
- Origin-exposure warnings
- Consequential-action classification
- Prompt-injection heuristics
- Exportable audit report
In development
- A desktop app for local protection, approvals, and activity review
- A bridge connecting compatible agents to WebMCP-enabled websites
- Security controls for discovery and actions across that connection
- Protection as agents move between websites and tools
- Clear approval for sensitive actions
- A useful record of what your agent did
Future
- Broader agent and browser compatibility
- More control for teams
- Stronger privacy choices
Guard will protect actions routed through its runtime. The Inspector is the available preview while that runtime is being built.
Public commitments
Reduces risk through layered controls.
Guard is designed to stay useful without quietly becoming another source of browser authority.
01Local-first analysis
02Minimal permissions requested only when needed
03No sale of browsing history
04No remote executable code
05No silent consequential execution
06Clear origin and agent identity
07Sensitive argument redaction in logs
08Configurable retention
09Signed, auditable releases
10Private browsing off by default