AI development tools
Features, installation, and usage notes for AI development tools
These tools bring code generation, retrieval, testing, and automation into software development workflows. Their permissions, supported environments, and maturity differ, so check official documentation and security boundaries before use.
Open Tools & Platforms
These open tools and platforms have public GitHub repositories; star counts are synchronized daily and dated. Code licenses, model licenses, and hosted-service terms are separate—consult the official repository and service documentation.
Claude Code
AI Programming Partner
Claude Code is Anthropic’s agentic coding tool for terminal and IDE workflows. It can inspect repositories, edit files, run commands and tests, and work with version-control workflows under user-configurable permissions.
Key features
- Understand entire codebases
- Write and modify code
- Run tests and builds
- Execute Git operations
- Use terminal commands
- Interactive programming conversation
Installation
Remote installer commands execute downloaded code. Review the official installation page and script before running them, especially in privileged or enterprise environments.
# Mac/Linux (recommended) curl -fsSL https://claude.ai/install.sh | bash # Windows (recommended) irm https://claude.ai/install.ps1 | iex # Homebrew (macOS/Linux) brew install --cask claude-code # WinGet (Windows) winget install Anthropic.ClaudeCode # Note: `npm install -g @anthropic-ai/claude-code` is deprecated
Usage
# Start an interactive session in any directory claude # Resume a previous session claude --continue # Update to the latest version claude update # Run a single prompt claude -p "explain the auth flow in this repo" # List built-in slash commands claude /help
MCP (Model Context Protocol)
Open Standard for AI Connectivity
MCP is an open standard protocol introduced by Anthropic to establish standardized connections between AI models and external tools/data sources. It’s called the "USB-C interface" for AI, enabling flexible interaction with various tools and data.
Key features
- Unified tool connection standard
- Support for multiple data sources
- Secure authorization mechanism
- Extensible architecture
Installation
Remote installer commands execute downloaded code. Review the official installation page and script before running them, especially in privileged or enterprise environments.
# MCP is itself a protocol — it is used through a client like Claude Desktop or Claude Code. # 1. Edit the config file: # macOS: ~/Library/Application Support/Claude/claude_desktop_config.json # Windows: %APPDATA%\Claude\claude_desktop_config.json # 2. Add an mcpServers block (see example below). # 3. Restart the client. # 4. Or install a reference server directly: npx -y @modelcontextprotocol/server-filesystem /path/to/folder
Usage
In Claude Desktop, add MCP servers through the settings menu to use them.
OpenAI Codex CLI
OpenAI’s Open-Source Terminal Coding Agent
openai/codex is a lightweight terminal coding agent released by OpenAI (github.com/openai/codex, written in Rust). It runs locally in your terminal, supports GPT-5 / o-series models, logs in via ChatGPT OAuth, and can load MCP servers. Codex CLI is the OpenAI counterpart to Claude Code.
Key features
- OpenAI-official open source
- Rust-based TUI
- Supports GPT-5 / o-series
- ChatGPT OAuth login
- Can load MCP servers
- Sandboxed code execution
Installation
Remote installer commands execute downloaded code. Review the official installation page and script before running them, especially in privileged or enterprise environments.
# Mac/Linux (recommended) curl -fsSL https://chatgpt.com/codex/install.sh | sh # Windows (recommended) powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex" # npm global npm install -g @openai/codex # Homebrew (macOS/Linux) brew install --cask codex
Usage
# Start an interactive session (auto OAuth login) codex # One-shot task (headless mode) codex exec "explain every .ts file under src/" # Switch model (Nous Portal / OpenRouter / self-hosted) # See: https://github.com/openai/codex
browser-use
AI Browser Automation
browser-use is the most popular AI browser-automation library on GitHub (github.com/browser-use/browser-use). It enables LLMs to operate real browsers like a human (click, type, scroll, screenshot), perfect for scenarios that need to interact with sites lacking APIs. Usable directly as a Python library or exposed as an MCP server to Claude Code and other clients.
Key features
- Supports all major LLMs (GPT / Claude / Gemini / local)
- Python library OR MCP server
- Real browser support (Chrome / Chromium)
- Vision + DOM dual-mode understanding
- Screenshots passed to LLM reasoning
Installation
pip install browser-use uvx --from 'browser-use[cli]' browser-use --mcp
Usage
# In a Python script:
from browser_use import Agent, ChatOpenAI
llm = ChatOpenAI(model="gpt-5")
agent = Agent(
task="Visit https://news.ycombinator.com and find top 5 AI stories today",
llm=llm,
)
result = await agent.run()
print(result)
# As MCP server:
# 1. Configure .mcp.json
# 2. Claude Code auto-loads on startupOpenCLAW
Open Source AI Assistant
OpenCLAW is an open source AI assistant project, meaning "The lobster way". It can run on any OS and platform as your personal AI assistant. This is a fully open source solution, allowing everyone to have their own AI assistant.
Key features
- Fully open source
- Cross-platform support
- Self-hostable deployment
- Active community
- Multiple model providers (selected during onboarding or configuration)
Installation
# Runtime requirement: Node.js 24 (recommended) or Node.js 22.19+ # Global install (npm) npm install -g openclaw@latest # Or via pnpm pnpm add -g openclaw@latest # Run the onboarding wizard (installs daemon as user service) openclaw onboard --install-daemon # Check gateway status openclaw gateway status
Usage
# 1. Run the wizard first openclaw onboard --install-daemon # 2. Send a test message openclaw gateway status # 3. Optional: start in foreground (debug mode) openclaw gateway stop openclaw gateway --port 18789 --verbose # 4. Docs: https://docs.openclaw.ai
Hermes-Agent
Growing AI Agent
Hermes-Agent is "The agent that grows with you"—an AI Agent that continuously grows with use. It can autonomously learn user habits, optimize task execution, and provide increasingly personalized services.
Key features
- Adaptive learning
- Long-term memory
- Personalized service
- Continuous task optimization
- Multi-modal support
Installation
Remote installer commands execute downloaded code. Review the official installation page and script before running them, especially in privileged or enterprise environments.
# Runtime: Python 3.11+ (handled automatically by installer) # Mac/Linux (one-liner) curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash # Windows (PowerShell) iex (irm https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.ps1) # The installer handles uv, Python 3.11, Node.js, ripgrep, ffmpeg, and a portable Git Bash
Usage
# 1. Run the setup wizard (configures model + tools) hermes setup # 2. Or: sign in via Nous Portal OAuth (one command) hermes setup --portal # 3. Start the terminal UI hermes # 4. Or: run the gateway and talk to it from Telegram/Discord/Slack hermes gateway setup hermes gateway start # 5. Switch model (OpenRouter has 200+ models) hermes model # Docs: https://hermes-agent.nousresearch.com/docs/
Anthropic Skills
Official Claude Skills Implementation Repository
anthropics/skills is Anthropic’s official reference implementation repository for Claude Skills — folders of instructions, scripts, and resources that Claude can dynamically load to perform specialized tasks. Skills teach Claude repeatable workflows: applying brand guidelines to documents, analyzing data with your org’s conventions, automating personal tasks, and more.
Key features
- Anthropic-maintained
- Dynamically-loadable Skills spec
- Skills API integration example
- Covers docs, data, and workflow domains
- Custom-Skill templates
- Follows agentskills.io standard
Installation
/plugin marketplace add anthropics/skills /plugin install document-skills@anthropic-agent-skills
Usage
Claude Code: mention an installed Skill by name in your request. Claude API: execute Skills through the Messages API `container.skills` field, with the documented beta headers and code-execution tool. GET /v1/skills lists Skills; it does not execute one. https://platform.claude.com/docs/en/build-with-claude/skills-guide
Claude Code Security Review
Anthropic’s Official PR Security Scanner (GitHub Action)
claude-code-security-review is Anthropic’s official GitHub Action (github.com/anthropics/claude-code-security-review) that uses Claude to automatically review PR diffs for security vulnerabilities. It supports diff-aware scanning (only analyzing changed files), posts findings as line-anchored PR review comments, and ships a built-in False Positive filter that automatically excludes DoS, rate-limiting, memory/CPU exhaustion, generic input validation, and open-redirect categories.
Key features
- Anthropic-maintained
- Diff-aware scanning (changed files only)
- Line-anchored PR review comments
- FP filter for 5 noisy categories
- Native GitHub Action integration
- Configurable Claude model (default Opus 4.1)
Installation
# Create .github/workflows/security-review.yml
name: Claude Code Security Review
on:
pull_request:
types: [opened, synchronize, reopened]
jobs:
security-review:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
with:
fetch-depth: 1
- uses: anthropics/claude-code-security-review@v0
env:
ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
# ⚠️ This Action is NOT hardened against prompt injection.
# Recommended to use only on trusted PRs, and pair with GitHub’s
# "Require approval for all external contributors" setting.Usage
# 1. Add ANTHROPIC_API_KEY secret to the repo # 2. On PR open/update, Action auto-analyzes diffs and posts # security findings as review comments # 3. Maintainers see risk severity + line location in the PR conversation
Ollama
Local Model Runtime
Ollama downloads and runs models on macOS, Linux, and Windows, with a CLI and local HTTP API. Whether a workload stays entirely local, how many resources it uses, and whether data reaches a remote service depend on the selected model and configuration.
Key features
- Local model download and execution
- Model library and Modelfile
- CLI and local API
- Embeddings, vision, and tool calling
- Python and JavaScript libraries
- Optional cloud models
Installation
Remote installer commands execute downloaded code. Review the official installation page and script before running them, especially in privileged or enterprise environments.
# macOS / Windows # Download the installer from https://ollama.com/download # Linux (inspect the remote script before running it) curl -fsSL https://ollama.com/install.sh | sh
Usage
# Start a model chat ollama run gemma4 # Confirm the installed version ollama -v
Hugging Face Hub
Open AI Collaboration Platform
Hugging Face Hub is an AI platform for discovering, sharing, and collaborating on models, datasets, and Spaces, with public and private resources. The huggingface_hub library and hf CLI support search, download, upload, cache management, and inference services; review each model’s and dataset’s license separately.
Key features
- Models, datasets, and Spaces
- Public and private repositories
- Model cards, dataset cards, and license metadata
hfCLI and Python library- Inference Providers and Endpoints
- MCP and Skill integrations for AI agents
Installation
# Install the Hugging Face Hub CLI and Python library pip install -U "huggingface_hub" # Log in only when private access or uploads are needed hf auth login
Usage
# Download one file from a public model repository hf download gpt2 config.json # Search models and limit the results hf models ls --search bert --limit 5
Abstract Concepts
AI Agent, Skill, and Harness are abstract concepts this site uses to organize the field, with no single GitHub repository. They describe patterns that span implementations rather than features of one product.
AI Agent
AI Systems That Execute Tasks Autonomously
An AI Agent is a system that plans and carries out complex tasks on its own. Unlike question-and-answer interaction, an Agent interprets a goal, forms a plan, calls tools, evaluates results, and adjusts its behavior from feedback. The term names a general pattern rather than one product: Claude Code, Codex CLI, Cursor, OpenClaw and Hermes Agent on this page are all concrete implementations of it.
Key features
- Autonomous task planning
- Tool usage capability
- Result evaluation and feedback
- Multi-step complex task handling
- Codebase understanding and operation
Installation
An Agent is a concept, not a separately installable product. To run one, install a concrete implementation; every entry under Open Tools & Platforms on this page gives its own installation steps.
Usage
Invocation differs by implementation: a command-line Harness usually accepts a described task directly, while framework-style implementations define the goal and the available tools in code. See the usage notes on the corresponding entry.
Skill
Ways to Extend AI Capabilities
A Skill packages domain knowledge, procedures, and reusable prompts into a module that loads on demand. A Skill is typically a directory holding a SKILL.md whose name and description decide when it is invoked. The format originated at Anthropic and is published, so it is not confined to a single client.
Key features
- Custom commands
- Workflow automation
- Domain-specific knowledge
- Reusable prompt templates
Installation
Place a directory containing a SKILL.md under ~/.claude/skills/ for personal use, or the project’s .claude/skills/. Skills can also be loaded through the Claude API’s Skills endpoint.
Usage
A Skill is usually matched and invoked automatically from its description; it can also be triggered explicitly as /skill-name.
Harness (Agent Scaffold)
The Runtime Host Layer for Agents
A Harness is the runtime layer that wires system prompt, tool loop, memory and plugins into a runnable agent. Claude Code, Codex CLI, Cursor Agent, and OpenClaw are all concrete Harness implementations — editing the Harness config is like rewiring the host, writing a new Skill is like plugging a new tool into the host. See the MCP page for how Harness fits with MCP/Agent/Skill.
Key features
- System prompt assembly
- Tool loop management
- Session state and long-term memory
- Skill and MCP scheduling
- Permissions and sandboxing
- Cross-form hosting (CLI/IDE/serverless)
Installation
Remote installer commands execute downloaded code. Review the official installation page and script before running them, especially in privileged or enterprise environments.
A Harness itself is a pattern, not a separately-installable product. To "use a Harness", install one of its implementations: # Claude Code (Node 18+) curl -fsSL https://claude.ai/install.sh | bash # Codex CLI curl -fsSL https://chatgpt.com/codex/install.sh | sh # OpenClaw (Node 24+) npm install -g openclaw@latest
Usage
Claude Code / Codex / Cursor / OpenClaw / Hermes Agent on this page are all Harness instances. Each has its own startup command (see the "Usage" section of each entry). You can also write a minimal Harness yourself: assemble a system prompt + tool loop + load MCP/Skill.
Closed-Source Products
Closed-source commercial products with no public GitHub repo to track.
Cursor
AI-Powered Code Editor
Cursor is an AI-enhanced code editor based on VS Code, with built-in AI capabilities similar to GPT-4. It can intelligently complete code, generate functions, explain code, and fix bugs. It is currently one of the most popular AI programming tools.
Key features
- Intelligent code completion
- Generate entire functions
- Code explanation and refactoring
- Chat mode interaction
- Built-in terminal
Installation
# Download from official site (recommended) # https://cursor.com # Or via WinGet (Windows) winget install Cursor.Cursor # Or via Homebrew (macOS) brew install --cask cursor
Usage
# 1. Launch Cursor # 2. Open a project folder # 3. Use Cmd+K (inline edit) or Cmd+L (chat) # 4. For Agent mode: Cmd+I to open Composer # 5. Sign in with your account to activate Pro/Ultra features
Conclusion
These tools can change coding, review, and automation workflows, but results depend on the task, model, and configuration. Before adoption, assess permissions, data handling, maintenance, and your team’s review requirements.