Plugins

GitHub

The Official Plugins Library is the official extension plugin library for Polaris AGI, fully compatible with and supporting the following mainstream industrial plugin protocols:

Anthropic MCP (Model Context Protocol) — stdio JSON-RPC 2.0 (2024-11-05 Version)
OpenAI Codex Plugin — Current standard protocol plugin

Core Plugins Library

1. Computer Use

Capabilities: Screenshots, mouse move/click/drag, keyboard input.
Tech Stack: mss, pynput, pywinauto

Native & Cross-Platform: Written in pure Python, natively calls OS underlying APIs (supports macOS / Windows / Linux).
Robustness & Hygiene: Built-in state cache TTL, macro-level anti-focus hijacking mechanism, and automatic screenshot garbage collection to ensure zero disk leakage.
Security: Strict App black/whitelist interceptor based on config.json.
Macro Adapter: Dedicated automation processor to bypass complex GUI restrictions (e.g. clipboard injection, intelligent UI OCR drift recovery).

2. Browser Use

Capabilities: Navigate URLs, click page elements, fill forms, capture screenshots.
Tech Stack: Python & uv package manager

Highlights: Allows AI Agents to directly drive and automate browser control via pure Python scripts for scraping, web operations, or complex information retrieval.

3. Knowledge Base

Capabilities: List directory contents, read file contents for RAG generation.
Tech Stack: Python os and pathlib native modules

Highlights: No external dependencies; supports sandbox isolation for read paths via the POLARISAGI_KB_ALLOWED_DIR environment variable.

Installation & Configuration

Step 1: Prerequisites

Please ensure Python 3.10+ and uv (blazing fast Python package manager by Astral) are installed on your system.

Step 2: Configure Plugins in Agent

For Claude Code / Claude Desktop

Add the following MCP Server config to ~/.claude.json:

{
  "mcpServers": {
    "polarisagi-computer-use": {
      "command": "uv",
      "args": ["run", "/absolute/path/to/polarisagi-plugins-official/plugins/computer_use/src/main.py"]
    },
    "polarisagi-browser-use": {
      "command": "uv",
      "args": ["run", "/absolute/path/to/polarisagi-plugins-official/plugins/browser_use/src/main.py"]
    },
    "polarisagi-knowledge-base": {
      "command": "uv",
      "args": ["run", "/absolute/path/to/polarisagi-plugins-official/plugins/knowledge_base/src/main.py"],
      "env": { "POLARISAGI_KB_ALLOWED_DIR": "/your/allowed/dir" }
    }
  }
}

For PolarisAGI AI Agent

PolarisAGI will automatically discover and install plugins from this repository via pkg/extensions/marketplace/, achieving an out-of-the-box ecosystem.

Repository Structure

plugins/
  computer_use/
    .codex-plugin/plugin.json   # Codex plugin manifest
    .mcp.json                    # MCP Server config
    src/main.py                  # Python MCP Server entry
    pyproject.toml
  browser_use/
    .codex-plugin/plugin.json
    .mcp.json
    src/main.py
    pyproject.toml
  knowledge_base/
    .codex-plugin/plugin.json
    .mcp.json
    src/main.py
    pyproject.toml

.agents/plugins/
  marketplace.json               # Global marketplace directory file