Tools for Agent Builders

Anthropic's agentic coding CLI. Reads your repo, edits files, runs commands, and ships features end-to-end from the terminal. Supports skills, MCP servers, hooks, and subagents.

OpenAI's open-source coding agent for the terminal. Plans, edits, and runs code in a sandbox with configurable approval modes. Reads AGENTS.md natively.

xAI's terminal coding agent. Plan Mode writes a file-by-file plan before editing, runs up to 8 parallel subagents in isolated git worktrees, and has native MCP support. Reads AGENTS.md and runs headless in CI. Powered by grok-build-0.1.

Open-source terminal coding agent for the Grok API, with live X and web search built in and MCP server support. A lightweight, BYO-API-key alternative to the official Grok Build.

Google's open-source terminal agent powered by Gemini. Large context window, built-in web search grounding, and MCP support. Generous free tier.

Pioneering terminal-based AI pair programmer. Works directly in your repo with tight Git integration, multi-file edits, and support for many model providers.

Open-source terminal coding agent with a polished TUI. Provider-agnostic — bring Claude, GPT, Gemini, or local models — with an LSP-powered understanding of your code.

Block's open-source AI agent for engineering tasks. Runs locally as a CLI or desktop app, works with any LLM provider, and is built around MCP extensions.

Open-source AI code assistant for VS Code and JetBrains. Build and share custom agents, models, and rules directly inside your editor.

Autonomous coding agent that lives in VS Code. Uses tools, edits files, runs commands, and works through complex tasks with your approval at each step. Roo Code is a popular fork with extra modes.

Low-level orchestration framework for building stateful agents as graphs. Durable execution, human-in-the-loop interrupts, and streaming — the workhorse for custom agent backends.

OpenAI's lightweight framework for multi-agent workflows: agents, handoffs, guardrails, and tracing with a small set of primitives. Works with 100+ non-OpenAI models too.

The standard TypeScript toolkit for AI apps: unified provider API, streaming UI, tool calling, and agent loops. The fastest path from idea to AI feature in a web app.

Python framework for role-based multi-agent teams ("crews") with sequential or hierarchical processes. Popular for getting collaborative agent workflows running quickly.

Extremely fast Python package manager from Astral. Its uvx command runs Python MCP servers (fetch, time, git…) with zero setup — a quiet prerequisite for half the MCP ecosystem.

Run open-weight models (Llama, Qwen, Gemma, Mistral, and more) locally with one command. The easiest way to give agents a private, offline model.

Call 100+ LLM providers through one OpenAI-compatible interface. Adds routing, fallbacks, budgets, and caching — the standard glue layer for multi-model agent systems.

Open-source LLM engineering platform. Tracing, evaluation, prompt management, and cost tracking for production agent systems. Self-host or use the cloud.

Open-source observability for AI agents and LLM applications. OpenTelemetry-based tracing, evaluation, and experimentation UI you can run locally.

Test and evaluate prompts, agents, and RAG pipelines like you test code. Side-by-side evals, CI integration, and red-teaming for LLM apps.

Official development tool for testing and debugging Model Context Protocol servers. Essential when building or integrating MCP servers.

All-in-one local RAG and agent workspace. Chat with your documents and codebases completely offline — great for giving agents private context.