Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

jev-guard-mcp

Experimental browser-only MCP server where Jev picks a code-owned action ID and a human approves before execution.

MCP serverprototype
View on GitHub ↗

Overview

Jev Guard MCP is an experimental, browser-only MCP server that lets Jev choose among browser actions owned by code without giving the model control of the browser. Codex supplies the intent, Playwright observes an isolated browser context separate from the user's Chrome profile and secrets, a policy engine reduces the page to bounded candidates (for example safe same-origin links), Jev selects one candidate ID, and the trusted MCP client shows the proposed action and obtains human approval before execution.

Before executing, the page is re-observed for freshness, only a fresh single-use preview token is accepted, and the postcondition is verified. The model never emits selectors, coordinates, JavaScript or arbitrary URLs. A public navigation guard stays restricted, and a separate supervised mode handles forms. The README notes the server cannot itself prove that the client obtained human approval. A Portuguese summary is included.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich code-generated candidate action best serves the goal?—

How it works

  1. Codex provides a URL and goal
  2. Playwright opens an isolated page; policy builds safe candidates
  3. Jev picks a candidate ID; client shows it for human approval
  4. Re-observe, execute the exact approved action and verify

Strengths

  • Model cannot produce arbitrary URLs or selectors
  • Isolated browser context
  • Human approval of each action

Caveats

  • Experimental
  • Approval relies on the MCP client

At a glance

Type
MCP server
Status
prototype
Language
TypeScript
Runtime
MCP server with Playwright
Has code
yes

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does Jev decide?

Which code-owned browser action to propose.

Can Jev navigate anywhere?

No, it only picks among candidates the policy engine generates.

Who approves actions?

A human, through the trusted MCP client.

Related projects

  • AskJev

    Claude Desktop MCP extension that drives Brave or Chrome with Jev choosing on-page actions and a guard freezing irreversible clicks.

    MCP serverbetaTypeScript
  • BrowserPaw

    Chrome MCP server with 50 tools that drives your everyday browser without losing logins, where Jev or a local decider picks each step.

    MCP serverbeta
  • jev-playwright-mcp

    Drop-in proxy for Playwright MCP that adds Jev page triage, prompt-injection masking, snapshot pruning and risky-action gating.

    MCP serverbetaTypeScript
  • jev-browser

    Library, CLI and MCP server where an LLM states each step's outcome and Jev decides element, action, value and done/blocked/irreversible.

    MCP serverbetaTypeScript
  • actiongate-jev

    Gateway that checks agent tool calls with policy plus Jev evidence, then binds approval to a single-use permit for that exact action.

    MCP serverbetaTypeScript
  • claude-jev (darwintechlab)

    OpenJev for Claude Code: plugin and MCP server giving Claude jev_choice, jev_noul, jev_score and jev_ask tools on the live API.

    MCP serverbetaTypeScript