The repo shelf: open source worth starring
Stars on GitHub measure popularity.
These are repositories we would actually point a friend to, grouped by what you are trying to do.
Connect your AI to things
modelcontextprotocol/serversReference MCP servers from the protocol project: filesystem, fetch, git, memory, and more. The best place to learn how servers are built.
github/github-mcp-serverGitHub's official server for repos, issues, pull requests, code search, and Actions.
microsoft/playwright-mcpLets an agent drive a real browser through the page's accessibility tree, which is faster and steadier than working from screenshots.
upstash/context7Feeds coding agents current, version-specific documentation so they stop writing code for last year's API.
Teach your AI a process
agentskills/agentskillsThe open Agent Skills specification and the place where the format is discussed.
anthropics/skillsExample skills from Anthropic, including document workflows, to study and adapt.
Run models yourself
ollama/ollamaThe simplest way to pull and run open models from the command line, with a local API other apps can use.
ggml-org/llama.cppThe inference engine behind much of local AI, and the source of the GGUF format.
open-webui/open-webuiA polished, self-hosted chat interface for local and remote models.
Make images and video
Comfy-Org/ComfyUIA node-based interface for building image and video generation pipelines with open models. Steep at first, unmatched for control once it clicks.
How we judge a repo before installing it
Recent commits, and maintainers who answer issues.
A clear license that fits how you plan to use it.
An install step you have read before you run it, especially anything piped into your shell.
For MCP servers: who maintains it, and exactly what access it asks for.
Do this today
Star the four repositories in the first group. Read the README for the reference servers from top to bottom. It is the fastest MCP education available, and it is free.
Resources
Every link this breakdown leans on, in one place.
Swipe