AI agents call browser_evaluate to retrieve information from Mcp without modifying anything — typically the context-gathering step in research, monitoring, and reporting workflows, before the agent takes action elsewhere.
While JavaScript evaluation could theoretically execute side effects, the tool's stated purpose is evaluation and return of results. However, JavaScript execution in a browser context carries inherent risks: it could access sensitive data (credentials, tokens, personal info), exfiltrate DOM content, or trigger unintended page behaviors. The 'serializable result' return suggests data extraction.
From the tool's definition Tool performs 'Evaluate a JavaScript expression in the page context and return a serializable result'—the action is introspective query/inspection of page state, not modification or execution of external operations.
Attacks that exploit this kind of access
Evaluate a JavaScript expression in the page context and return a serializable result. It is categorised as a Read tool in the Mcp MCP Server, which means it retrieves data without modifying state.
Register the MCP server in PolicyLayer and add a rule for browser_evaluate: allow, deny, rate-limit, or require approval. Point your MCP client at the PolicyLayer proxy URL and the rule is enforced on every call, before it reaches Mcp. Nothing to install.
browser_evaluate is a Read tool with low risk. Read-only tools are generally safe to allow by default.
Yes. Add a rate_limit block to the browser_evaluate rule in your PolicyLayer policy. For example, setting max: 10 and window: 60 limits the tool to 10 calls per minute. Rate limits are tracked per agent session and reset automatically.
Set action: deny in the PolicyLayer policy for browser_evaluate. The AI agent will receive a policy violation error and cannot call the tool. You can also include a reason field to explain why the tool is blocked.
browser_evaluate is provided by the MCP server (victormyschik/mcp). PolicyLayer sits as a proxy in front of this server to enforce policies before tool calls reach the server.
Every MCP server has a record like this.
Type a name, get the same breakdown: verified identity, auth posture, risk grade, capabilities, recommended policy.
Teams ship this data inside their own products. See what a licence covers →