AI that matters, from the architect's desk. Curated and engineered by Saaket Varma, PhD — no hype, just signal.
HOW TO READ THIS Read top to bottom: Google ships the CLI, its terminal command reaches Google Cloud, which provisions agents, driving 4.7K GitHub stars.
Google released agents-cli, a command-line tool that turns any coding assistant into a builder, evaluator, and deployer of AI agents on Google Cloud. This isn't a wrapper around someone else's framework — it's Google staking out first-party ground in agent tooling, the same way AWS and Anthropic have been doing with their own stacks. If you're building agentic workflows on GCP, this collapses a chunk of the plumbing you'd otherwise hand-roll: creation, eval, and deployment in one CLI. Worth a look if Google Cloud is anywhere in your stack — it's the clearest signal yet that agent infrastructure is becoming a first-party platform bet, not a side project.
HOW TO READ THIS Read top to bottom: the DevTools team builds it, ships it as an MCP server, an agent then drives the browser through it, and the repo lands 46K stars.
Chrome's own DevTools team shipped an MCP server giving coding agents direct control over DevTools — inspecting, debugging, and manipulating pages without scraped or fragile browser hacks. This matters because browser control has been a persistent weak point in agent tooling; having the actual Chrome team own the primitive means better fidelity and longer support life than any third-party bridge. If your agents touch web UIs or need to debug frontend issues, this is the one to standardize on.
HOW TO READ THIS Read top to bottom: an agent links through the MCP bridge into Unity's editor, which branches into three exposed surfaces, driving 12K GitHub stars.
Unity MCP bridges AI assistants directly into the Unity Editor — managing assets, controlling scenes, and editing scripts from an LLM. It extends MCP well beyond coding tools into game development, reinforcing that MCP is becoming the default connective tissue for agent-to-application control, not just agent-to-filesystem. Worth watching if you're anywhere near game dev or real-time 3D tooling — this is the pattern other creative-software vendors will likely copy next.
HOW TO READ THIS Read top to bottom: the builder ships a repo, that repo adds a /watch command, the command streams video frames into Claude, and the tool's stars surge to 3.6K.
A new /watch tool lets Claude download a video, extract frames, transcribe the audio, and reason over the whole thing as one artifact. It's a community tool, not an official Anthropic release, but it's a useful preview of where multimodal agent tooling is heading — past text and code into raw video understanding. If you work with video content pipelines, this is a cheap way to prototype video-aware agent workflows today.
Multi-agent LLM framework for financial trading — agents specialized by role (analyst, trader, risk manager) debating a trade decision.
Microsoft's official skills repo for AI coding agents working in .NET/C# — first-party agent tooling from the .NET team itself.
Terminal AI coding agent with hash-anchored edits, LSP integration, and subagent support — a leaner alternative to the big CLI agents.
Visual annotation tool for reviewing coding-agent plans and diffs, with one-click feedback routed straight back to the agent.
Menu-bar usage tracker for OpenAI Codex and Claude Code — no login required, just local usage stats at a glance.