AI AGENTS & MCP · CURATED DIRECTORY
AI agents are programs that break down goals, call tools, and complete tasks autonomously — from writing code to booking flights, they are moving from chatting to doing. MCP (Model Context Protocol) is the open standard connecting AI to the outside world, letting one model safely use files, browsers, GitHub, and more.
POPULAR AI AGENTS
From closed-source flagships to open-source frameworks, covering coding, general execution, and multi-agent collaboration. Star counts are approximate.
Anthropic's terminal-native coding agent that reads and edits your codebase right in the CLI.
A general-purpose AI agent from Butterfly Effect that autonomously completes full task chains, from research to delivery.
Cognition's AI software engineer that independently plans, codes, and ships entire features.
The open-source pioneer that let GPT-4 decompose goals and loop on tasks — the project that ignited the agent boom.
Assemble autonomous AI agents right in your browser — no code needed, and deploy to the cloud.
A role-playing multi-agent framework where agents with distinct roles collaborate like a team on complex workflows.
LangChain's low-latency agent orchestration framework — build controllable, recoverable agent state machines with graphs.
A multi-agent framework powered by SOPs — it runs like a software company, producing high-quality deliverables.
MCP SERVERS · OFFICIAL
Official reference implementations maintained by Anthropic in the modelcontextprotocol/servers repo. Give any AI model these capabilities — it is like growing it a pair of hands.
Give AI safe read/write access to local files, with project browsing and directory listing.
Wraps the GitHub API so agents can create issues, open pull requests, and manage repositories directly.
Knowledge-graph-based persistent memory so agents remember facts and relations across sessions.
HOW TO CHOOSE
Coding, writing, research, or automation? Different agents excel at different jobs — pick Claude Code or Devin for programming, Manus for general execution, CrewAI for research collaboration. Define the core scenario before choosing.
Sensitive codebase or business data? Prefer self-hostable open-source options (AutoGPT, CrewAI, MetaGPT); choose commercial products for out-of-the-box stability. Keep data where you can control it.
Don't build a complex multi-agent pipeline on day one. Run one agent on a single task first, verify output quality and cost, then scale up. You only learn an agent's limits by using it for real.
Search thousands of AI tools on our site, or browse the skills library to see agents in action.