A Builder's Reading List
This isn't a "top 10 GitHub repos" listicle. These are the specific repositories that power my client work and stay in my rotation. I check commits on most of these weekly. When something shifts in the Claude ecosystem or MCP landscape, these repos are where I find out first.
Updated April 2026 · 10 repositories
The Foundation
These three repos form the backbone of how I build AI systems. If you're serious about shipping with Claude and Next.js, start here.
This is where I go before building any new Claude integration. RAG patterns, tool use, sub-agents. Working code, not documentation. I pull from it constantly.
The reference implementation library for everything Claude API. When you're building a new integration and need to see how Anthropic themselves would approach it (RAG, tool use, prompt caching, sub-agents), this is where you look first. Not documentation: actual working code. Pull from it constantly.
View on GitHubBefore writing any MCP server from scratch, I check what already exists here. The official implementations save hours and set the quality bar for everything else.
The canonical source for official MCP server implementations: filesystem, GitHub, PostgreSQL, Brave Search, and more. When building a new MCP integration, start here before writing anything from scratch. Reference implementations save hours. Managed by Anthropic.
View on GitHubThe AI SDK that powers every production app I build. Streaming, multi-model support, tool use. If you're shipping AI features in Next.js, this is the foundation.
The AI SDK for TypeScript, from the team that builds Next.js and Vercel. Streaming responses, multi-model support, tool use, and generative UI all handled cleanly with a unified interface. If you're building production Next.js apps with AI features, this is the toolkit to build on rather than rolling your own stream handlers.
View on GitHubAnthropic orbit
Runs Claude Code inside GitHub Actions: PR reviews, automated test generation, commits on merge. The official CI/CD integration for Claude Code. This one is moving fast. Watch it so you know when it's ready for production use.
MCP ecosystem
Community-maintained directory of 1,200+ MCP servers organized by category: automation, cloud, browser, code execution, databases, and more. This is how you find an existing MCP capability before building your own. Check it every time you're about to build something new. Updated daily.
The cleanest MCP-native agent framework available. Implements every pattern from Anthropic's Building Effective Agents (orchestrator, evaluator-optimizer, router, parallel fan-out) in composable, production-ready code. Before writing custom orchestration logic for any agent workflow, read this repo first. Scales to Temporal for durable, resumable agents.
The companion to awesome-mcp-servers. A curated list of MCP clients: Python frameworks, desktop chatbots, VS Code extensions, CLI tools, and agentic code editors. Useful for understanding what client implementations exist before building your own, and for seeing how other tools connect to MCP servers you've built.
Infrastructure
LLM gateway that calls 100+ models in unified OpenAI format, with cost tracking, load balancing, fallbacks, and logging. Directly relevant to how an AI gateway architecture works. Also your insurance policy: if pricing, rate limits, or capabilities shift on any model, you swap one config line instead of rewriting integrations.
Self-hosted workflow automation with 400+ integrations and native AI capabilities, including Claude. The power-user alternative to Make.com: you own the data, you control the pricing, you control the deployment. As automation stack complexity and client data sensitivity increases, this is where serious operators land.
Reference
The most comprehensive prompt engineering reference that exists, continuously updated as new techniques emerge. Not a blog post: an actual living reference covering chain-of-thought, RAG, agentic patterns, and multi-modal prompting across all major models. Watch the commits. When something new gains traction, it shows up here within days.
Building with these tools? I build custom AI systems for B2B mid-market companies. Learn about Custom AI Builds →
See the full tech stack
Tech Stack →Frequently Asked Questions
What are the best GitHub repos for learning Claude and MCP?
The anthropic-cookbook is the official reference implementation library for everything Claude API. The modelcontextprotocol/servers repo has canonical MCP server implementations. Together they cover API integration patterns and the MCP ecosystem.
What GitHub repos should an AI developer follow?
Start with vercel/ai for the TypeScript AI SDK, then punkpeye/awesome-mcp-servers for the community MCP directory. Add BerriAI/litellm for multi-model gateway architecture and n8n-io/n8n for workflow automation with AI capabilities.
What is the best prompt engineering resource on GitHub?
The dair-ai/Prompt-Engineering-Guide with over 53,000 stars is the most comprehensive reference available. It covers chain-of-thought, RAG, agentic patterns, and multi-modal prompting across all major models. Continuously updated as new techniques emerge.