Tools and skills for AI agents

The de-facto standard for building LLM applications: ready abstractions for chained calls, tool wiring, conversation memory, and integrations with dozens of data sources. Its huge ecosystem and community mean a connector or example exists for almost anything. Use it to prototype an agent or RAG pipeline fast without building infrastructure from scratch.

A library from the LangChain team for describing agents as state graphs: nodes, edges, cycles, and conditional transitions give precise control over how an agent reasons and when it stops. It supports state persistence, human-in-the-loop, and fault tolerance. You need it when a simple chain is no longer enough and you want a reliable, reproducible production workflow with branching and retries.

A framework for orchestrating 'crews' of agents: you define roles (researcher, analyst, writer), goals, and tools, and CrewAI distributes tasks and coordinates their collaboration. Its strength is an intuitive role-and-process model close to how real teams work. It fits tasks that naturally split into stages and specializations — research → analysis → report.

Microsoft's research framework for multi-agent systems: several agents converse, delegate subtasks, and jointly reach a solution, including executing code. A flexible event-driven architecture supports both simple and complex agent configurations. Use it when a task is best solved through a dialogue of specialized agents rather than one monolithic prompt.

A lightweight, fast framework for agents and multi-agent systems with built-in memory, a knowledge base, and out-of-the-box tool support. It emphasizes minimal overhead and fast agent instantiation. It fits when you need to spin up many agents or handle high load without sacrificing memory and knowledge access.

An open-source RAG engine focused on deep document parsing: it carefully parses PDFs, tables, scans, and complex layouts, then answers questions with links to the specific source fragments. It ships as a ready service with a web UI. Use it when you need reliable answers over a large, varied document base and want every answer verifiable.

A data framework specializing in connecting LLMs to your own data: loaders from hundreds of sources, different index types, advanced retrieval strategies, and building agents on top of your knowledge. It is especially strong at fine-tuning retrieval — from simple vector search to complex query graphs. Choose it when the heart of the app is high-quality search, not agent orchestration.

deepset's mature, production-oriented framework for search and RAG pipelines: composable nodes (retrievers, readers, generators) assemble into reliable pipelines with clear contracts between steps. It emphasizes stability, testability, and scaling in real services. Use it when a RAG system must move beyond prototyping into production and stay maintainable.

The official set of Model Context Protocol (MCP) reference servers — an open standard through which agents access files, databases, APIs, and other tools in a uniform way. These servers serve as both ready integrations and a template for writing your own. Use them as a starting point when you want to give an agent controlled access to external systems without hand-rolled connectors.

A large curated catalog of MCP servers: hundreds of ready integrations — databases, search, clouds, messengers, dev tools — with links and a short description of each. It is community-maintained and regularly updated. Look here to quickly find a ready MCP server for a specific system instead of writing an integration from scratch.

A platform and SDK giving agents 250+ ready tools and integrations with popular services (GitHub, Slack, Gmail, CRM, and dozens more) plus managed authentication. It removes the OAuth and API wrangling for each service. Use it when an agent needs to actually act in external systems, not just reason, and you don't want to write and maintain those integrations yourself.

A tool for systematically testing, evaluating, and red-teaming prompts and LLMs: you define test-case suites and metrics and compare how different models and phrasings handle a task. It helps catch quality regressions and vulnerabilities before release. Use it when you need to objectively choose a model or prompt by the numbers, not by feel, and keep quality under control as things change.

One of the most famous and most-starred collections of ready prompts for chat models: hundreds of roles and scenarios ('act as…') for any case — from writing code to marketing and teaching. A great starting point and source of ideas for your own system prompts. Look here to quickly find a working phrasing for a task instead of inventing one from scratch.

The best-known curated list for Claude Code: it collects working slash commands, CLAUDE.md files, hooks, ready workflows, and community integrations. A convenient entry point for adopting others' tricks and quickly leveling up your setup. Look here to get the most out of Claude Code without inventing commands and configs from scratch.

A large curated collection of ready subagents for Claude Code from VoltAgent: dozens of specialized roles — from code reviewer and DevOps to data analyst — with ready definitions and system prompts. Use it as a library to copy subagents straight into your project. Handy for quickly assembling a team of specialized agents instead of writing each by hand.

A multi-harness marketplace of agent plugins: one set of subagents, commands, and skills compatible with Claude Code, Codex CLI, Cursor, OpenCode, Copilot, and Gemini CLI at once. It's useful because it doesn't tie you to a single tool. Use it if you work with several CLI agents and want to reuse the same setup everywhere.

A catalog and CLI for configuring and monitoring Claude Code: ready project templates, subagents, commands, and MCP configs installable in a couple of clicks. It helps quickly bring a repository to an agent-friendly structure. It fits when you want to start from a proven configuration instead of setting everything up by hand.

A curated collection around Agent Skills: tutorials, guides, and catalogs of ready skills you can use to extend agents for specific tasks. Good as an overview map of the topic and a source of ideas. Look here to understand how agent skills work and find ready examples to reuse.

A huge, very active collection of ready open-source LLM, agent, and RAG applications: dozens of working examples with step-by-step READMEs — from chatbots to multi-agent systems. A great base for learning from real code and starting your own project fast. Use it as an architecture reference and a source of ready scaffolds for your task.