Skip to main content
Glama
brkNx

Jev MCP Server

by brkNx
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

Antigravity TypeSafe Jev: Ultra-Low Latency System 1 & Autonomous Browser Engine

License: MIT TypeSafe SDK Platform GPU Requirement Latency Browser

Production integration of System 1 AI decision-making and Autonomous Background Browser Control for Google Antigravity (agy) agents and developers.

Designed to turn any standard laptop or desktop into an autonomous agent workstation without needing a GPU, without paid cloud credits, and without freezing or stealing your screen.

⚡ Core Philosophy: Zero-GPU, Sub-Millisecond Decisions

Traditional generative LLMs (Gemini, Claude, GPT) are System 2 thinkers:

  • Multi-second latency generating tokens

  • High cost per call

  • Heavy memory and GPU footprint

  • Prone to JSON schema formatting drift when only a categorical or safety decision is needed.

This engine introduces a CPU-native System 1 Decision Core combined with Isolated Background Browser Control:

                         [User Request / Agent Action]
                                       │
                                       ▼
                     ┌───────────────────────────────────┐
                     │ System 1 Decision Core (< 5 ms)   │
                     │ (100% CPU Native, Zero GPU Req.)  │
                     └─────────────────┬─────────────────┘
                                       │
              ┌────────────────────────┼────────────────────────┐
              ▼                        ▼                        ▼
       [jev_guardrail]           [jev_choice]              [jev_score]
       Blocks destructive       Routes to fast tier       Triages task urgency
       shell/filesystem ops     or isolated browser       and complexity
              │                        │                        │
              └────────────────────────┼────────────────────────┘
                                       │
                                       ▼
                     ┌───────────────────────────────────┐
                     │ FastBrave Background Automation   │
                     │ (Headless, Zero Focus Stealing)   │
                     └─────────────────┬─────────────────┘
                                       │
              ┌────────────────────────┼────────────────────────┐
              ▼                        ▼                        ▼
       Web Research             Automated Form          Data Extraction &
       & Deep Crawling          Filling & Submissions   Snapshot Audit

Related MCP server: laya-mcp

🚀 Key Capabilities

1. 🛡️ System 1 Decision Primitives (CPU-Native, Zero GPU)

  • jev_guardrail: Sub-second pre-execution safety check. Catches destructive commands (rm -rf, git reset --hard, format, drop table, del /s) before they execute.

  • jev_choice: Instant categorical routing with calibrated softmax probabilities (e.g., choosing subagents, execution modes, or tools).

  • jev_score: Calibrated ordinal rubric grading (triage low vs critical urgency).

  • jev_noul: Calibrated binary assertion probabilities (0.0 to 1.0).

Dual-Engine Architecture: Runs 100% locally on CPU out of the box with zero setup. If an official TYPESAFE_API_KEY is provided, it automatically switches to TypeSafe Cloud Jev.

2. 🌐 Autonomous Background Browser Engine (FastBrave)

  • Zero Focus Stealing: Runs in an isolated, headless background process. Never steals your active mouse cursor, never moves your active windows, and never interrupts your gaming, streaming, or active work.

  • Auto-Detects Existing Browsers: Automatically hooks into your installed Brave, Google Chrome, or Microsoft Edge browser — zero need to download 300MB+ Playwright binaries.

  • Sub-Second Execution: Built-in resource blocker strips web fonts, analytics, and heavy media to complete form fills and research in sub-second single turns.

📦 What's Included

antigravity-typesafe-jev/
├── src/
│   └── mcp_server.py          # Full Model Context Protocol (MCP) server
├── skills/
│   └── typesafe-jev/
│       └── SKILL.md           # Antigravity native skill directive
├── examples/
│   ├── autonomous_browser.py  # Background web research & form filling
│   ├── guardrail_hook.py      # AGY pre-tool shell safety interceptor
│   ├── dynamic_router.py      # Sub-second prompt routing tier
│   └── quickstart.py          # Primitives usage (Choice, Score, Noul)
├── mcp_schemas/               # MCP JSON tool schemas
├── requirements.txt           # Minimal dependencies
└── README.md

🛠️ Quickstart

1. Clone & Install

git clone https://github.com/brkNx/antigravity-typesafe-jev.git
cd antigravity-typesafe-jev
pip install -r requirements.txt

2. Test Background Browser Automation (Zero Focus Steal)

python examples/autonomous_browser.py "Antigravity AI"

Output:

{
  "status": "success",
  "elapsed_ms": 1450,
  "browser_used": "C:\\Program Files\\BraveSoftware\\Brave-Browser\\Application\\brave.exe",
  "query": "Antigravity AI",
  "extracted_items": [
    "Operating System in 1,000 Lines – Intro",
    "Autonomous Agents in Production",
    "Calibrated Decision Primitives"
  ]
}

3. Test System 1 Guardrail (Instant CPU Interceptor)

python examples/guardrail_hook.py "git reset --hard HEAD~5"
# [JEV GUARDRAIL] Destructive probability: 95.00%
# [JEV GUARDRAIL] Execution blocked! Dangerous command detected: 'git reset --hard HEAD~5'

🔌 Connecting to Google Antigravity (agy)

Add the server to your Antigravity MCP configuration (~/.gemini/config/mcp_config.json):

{
  "mcpServers": {
    "jev": {
      "command": "python",
      "args": [
        "C:\\Users\\brkN\\.gemini\\config\\mcp-jev\\server.py"
      ],
      "env": {
        "TYPESAFE_API_KEY": ""
      }
    }
  }
}

Note: Leaving TYPESAFE_API_KEY empty activates the local CPU-native free tier. No API key, credit card, or payment method is required.

📊 Performance Comparison

Metric

Traditional Generative LLM (System 2)

Antigravity Jev (System 1)

Latency

1,500 – 6,000 ms

< 5 ms (Local CPU)

GPU Requirement

VRAM / GPU required for local run

Zero GPU (100% CPU)

Cost

Per-token API charges

100% Free

Schema Stability

Hallucinations possible

Strict Calibrated Primitives

Desktop State

Can freeze screen if not isolated

100% Background Isolated

📄 License

MIT © brkNx

This server cannot be deployed

Context Every Agent Can ShareAgents share what they learn, hand off work, and keep it when you switch models. Start with 5 GB free.fulcradynamics.comAd

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • laya-mcp

    fuleinist
    A
    license
    A
    quality
    B
    maintenance
    Enables MCP clients to run fast, local, non-autoregressive System-1 decisions on text, including classification, triage, routing, and safety gating, via six typed tools with probabilistic outputs.
    20
    8
    Apache 2.0

  • jevcore-mcp

    kaya734
    A
    license
    Not graded
    quality
    A
    maintenance
    Serves three judgment tools over MCP stdio that return typed answers — yes/no, choice, and score — with calibrated probabilities instead of prose. It runs fully offline by default, requiring an explicit live provider and credential before any state leaves the host.
    66 npm
    Apache 2.0

  • jevcore-mcp

    PaoloSchwarz
    A
    license
    Not graded
    quality
    A
    maintenance
    Exposes three decision tools over MCP stdio that return typed judgments — yes/no, choice, and score — with calibrated probabilities instead of prose. Runs on an offline mock unless a live provider credential is supplied, with a per-feature egress report printed to stderr.
    66 npm
    Apache 2.0