Five Tools That Actually Shipped This Week
What developers shipped this week.
Five Tools That Actually Shipped This Week
The Shiplog
awslabs/mcp - Amazon's official collection of MCP servers for AWS services. Written in Python with over 9,000 stars, this project connects Claude directly to S3, EC2, Lambda and other AWS primitives through the Model Context Protocol. Developers building agentic workflows that span cloud infrastructure should care: this removes the boilerplate of writing custom API wrappers. Worth your time if you are deploying to AWS and want AI tools that can actually interact with your infrastructure rather than just describing it.
tirth8205/code-review-graph - A local-first code intelligence graph that builds a persistent map of your codebase for AI-assisted review. It runs as an MCP server and CLI, meaning Claude can maintain context about your entire project across sessions. Most code review tools treat each session in isolation. This one remembers. Useful for teams maintaining large codebases where institutional knowledge about architecture lives in the code itself. Early stage project but the idea is sound.
Portkey-AI/gateway - An AI gateway written in TypeScript with integrated guardrails and support for routing to over 1,600 LLMs. With 11,862 stars it is the most-starred item in this week's pipeline by a significant margin. The guardrails component is what sets it apart: production teams who need to route requests across providers while enforcing content policies have a legitimate option here. Worth investigating if you are running multi-provider AI infrastructure.
zebbern/claude-code-guide - A community guide to Claude Code covering setup, commands, workflows, agents, skills and advanced patterns. At 4,168 stars it is the most popular Claude Code specific resource in this week's pipeline. The guide aggregates workflows that would otherwise require digging through documentation, Reddit threads and GitHub issues. Worth bookmarking even if you think you know Claude Code well.
su-kaka/gcli2api - A Python project that converts GeminiCLI and Antigravity tools into OpenAI, Gemini and Claude API interfaces. With 4,826 stars it solves a real problem: tools built for one AI ecosystem becoming portable to others. Developers who have invested in CLI workflows built for other providers can now expose those capabilities to Claude without rewriting them. Practical and under appreciated.
The Take
This week's pipeline reveals a pattern that should concern anyone building production AI systems: the tooling ecosystem is fragmenting faster than anyone can maintain it.
Look at what actually shipped. We have multiple MCP clients competing for attention (5ire, Klavis, Deepchat), several Claude Code reference resources (two separate cheat sheets and a guide), and proxy services trying to abstract across provider boundaries. Each project solves a real problem for its author. But a developer landing in this ecosystem today faces a coordination problem, not a capability problem. Every tool works in isolation. None of them talk to each other without friction.
The MCP specification was supposed to solve this. And it is helping. But standardisation of the protocol does not mean standardisation of the ecosystem. We are building toward a world where every AI tool speaks the same language but nobody has agreed on what to say.
The practical implication: teams investing in AI tooling today should be deliberate about which integration points they commit to. The proxy and gateway layer (Portkey, Claude Code Hub, gcli2api) exists precisely because integration points are proliferating faster than consolidation can keep pace. That gap will close eventually. But "eventually" is not much comfort when you are building something that needs to work next quarter.
Forward this to one person who should be using AI better than they are. Reply with what you built, tried or broke this week. I read every one.
Gareth, founder of The Anthropic Stack (theanthropicstack.com)