What makes a tool an "AI agent" instead of a chatbot?

Agency. A chatbot answers a prompt and stops; an agent takes a goal, plans steps, calls external tools (terminal, browser, files, APIs), checks its own results, and loops until the task is done or it needs your input. Everything on this page runs some form of that loop — a pure autocomplete or one-shot Q&A tool doesn't qualify.

Are open-source AI agents good enough for real work?

Yes — and for coding agents specifically, open source is arguably the mainstream. Tools like OpenCode and Kimi Code let you bring your own model (Claude, GPT, DeepSeek, or a local one), audit exactly what the agent does, and avoid vendor lock-in. The trade-off is setup effort and fewer guardrails than a managed product.

Should I pick a terminal agent or an IDE agent?

Match the tool to your workflow, not the hype. Terminal agents (Claude Code, OpenCode, Grok Build) are scriptable, work over SSH, and slot into CI. IDE agents (Cursor, Superset, CLODEx) give you visual diffs, parallel agents in git worktrees, and tighter editor context. Most are free to try — run both on a real task for a day and keep the one that sticks.

What is MCP and why does it keep showing up in these listings?

MCP (Model Context Protocol) is an open standard for connecting agents to external tools and data sources. An agent with MCP support can plug into databases, browsers, ticketing systems, or your own internal APIs through one common interface, so you're not limited to the integrations the vendor shipped. We flag MCP support because it usually means the agent is extensible rather than a walled garden.

Is it safe to let an AI agent loose on my codebase?

With the right defaults, yes. Run agents inside git so every change is a reviewable diff, start with read-only or approval-required modes, and keep secrets out of reach (most agents respect gitignore-style exclusions). Local-first and open-source agents make this easier to verify, since you can see exactly which commands run. Treat an agent like a fast junior colleague: review its work before you merge it.

AI Agent Tools Directory

37 agent tools, screened by an editor

AI tools are splitting into two camps: the ones that answer, and the ones that act. This page tracks the second camp — agents that take a goal, break it into steps, call real tools (a shell, a browser, an API), and keep iterating until the job is done or they need your call.

Agentic coding is where the shift is most visible, and it's the part of the ecosystem we know best — this site itself is built and maintained with tools like OpenCode, Superset, and Claude Code. Alongside the coding CLIs and IDEs you'll find the supporting cast: browser and research agents, agent memory layers, MCP connectors, and the routers and orchestration frameworks that tie it all together.

Every listing here is screened by an editor, not scraped. We favor tools you can actually run today — open-source and local-first projects sit next to commercial ones, and we note model support, pricing model, and platform so you can shortlist without opening fifteen tabs. The list is re-curated as the ecosystem moves; a tool that stops shipping gets dropped.

The directory (37 tools)

Featured