The Claude Code tools I actually use.
Every week I open a dozen new repos and try the ones that look like they solve a real problem. Most get closed. What is left is wired into my own setup as an MCP server, an installed skill, or a CLI I run.
No affiliate filler. The first two sections are wired into my setup. The last is what I am watching but not running yet.
Tools and MCP servers.
The standalone tools and MCP servers that handle context, memory, and search. These do the unglamorous work that keeps long sessions fast and cheap.
token-optimizer
Audits where your context actually goes and strips the dead weight. I run it when a session starts feeling sluggish and the token count is climbing for no reason.
View on GitHubqmd
A local search engine for your notes, docs, and meeting transcripts. It lives on your machine, so Claude can search thousands of files without burning context reading them one by one.
View on GitHubOpen Brain
A memory layer that every AI you use can plug into. One database, one place your context lives, instead of re-explaining yourself to a fresh model every morning.
View on GitHubpaperclip
Open-source orchestration for running a company with agents instead of headcount. Early and ambitious, and exactly where I think solo operations are heading.
View on GitHubfree-claude-code
Runs Claude Code in your terminal, VSCode, or Discord through cheaper or local backends. Worth a look if your usage bill is climbing faster than your revenue.
View on GitHubHyperlink
An AI sidekick that works across your local files. Ask it something and it pulls the answer from your own documents instead of the open web.
Visit siteKarpathy's LLM wiki
Andrej Karpathy's running notes on how LLMs actually work. Not a tool, a reference. Read it when you want the mental model behind what you are automating.
Read the wiki
Stop collecting tools. Build one agent.
In 60 minutes I walk you through building a Claude Code agent that runs one task you're sick of doing. Live on Zoom, $1 to get in the room.
Join the workshopClaude Code skills and plugins.
Skills change what Claude can do inside a session. These are the public ones I run. My own content and copywriting skills stay private, so they are not here.
superpowers
A library of process skills that change how Claude works, not just what it knows. Brainstorming, test-driven development, and debugging all kick in automatically when the task calls for them.
View on GitHubobsidian-skills
Built by the maker of Obsidian. It lets Claude read, write, and search my vault, work with Bases and Canvas files, and pull clean markdown out of any web page. If you live in Obsidian, this is the bridge.
View on GitHubCodex in Claude Code
Drops OpenAI's Codex inside Claude Code. When Claude gets stuck or I want a second opinion on a hard bug, I hand the job to Codex without leaving the terminal.
View on GitHublast30days
Pulls what people are actually saying about a topic in the last 30 days across Reddit, X, YouTube, TikTok, Hacker News, and the web. I run it before writing about a trend so I am working off real signal, not guesses.
View on GitHubn8n-mcp
Gives Claude the full n8n node catalogue and thousands of real workflow templates. This is what lets Claude build and validate working n8n automations from a plain-English description.
View on GitHubqwoted-seo-backlinks-skill
Turns Claude into a PR machine. It finds journalist requests on Qwoted, builds a sourced stats page worth citing, and drafts the pitch. This is how you earn backlinks from real publications instead of buying junk ones.
View on GitHub
Installing skills isn't shipping.
Bring one task you do every week. Leave the workshop with the Claude Code agent that does it for you. Live on Zoom, no replay, $1 to get in.
Join the workshopRecommended, on my radar.
Tools I rate but do not run daily yet. Open in my tabs, worth a look, and a few of these will graduate to the stack once I have put them through real work.
claude-video
Gives Claude eyes for video. /watch downloads a clip, pulls the frames, transcribes the audio, and hands the whole thing over. Useful when the answer you need is buried in a 40-minute YouTube video.
View on GitHubdistribb-skill
Write content with your own AI, then publish it through a backlink network from the command line. Pairs with the Qwoted skill if SEO is your game, and there is a free plan to start on.
View on GitHubcrush
Agentic coding in the terminal from the Charm team. Point it at any model provider you have a key for, which makes it a clean fallback when Anthropic is having a bad day.
View on GitHubhindsight (Vectorize)
Vectorize's agent memory that learns from what worked and what did not. Different angle from Open Brain. This one is about the agent getting better over time, not just remembering facts.
View on GitHubmempalace
Open-source AI memory, benchmarked against the paid options and free to run. On the list for the same job as hindsight, worth comparing the two.
View on GitHubagentscope
A framework for building agents you can actually watch run. You see what the agent is doing at each step instead of staring at a black box.
View on GitHubautoresearch
Andrej Karpathy's experiment in agents running their own research loops on a single GPU. More research toy than daily driver, but the ideas are ahead of the curve.
View on GitHubTaste Skill
An anti-slop frontend framework for agents. It stops your AI from shipping the same generic layout everyone else ships.
Visit sitemattpocock/skills
Matt Pocock's open collection of Claude skills. Good place to study how someone who teaches this stuff structures a skill.
View on GitHub
New to Claude Code?
Most of these assume you already have Claude Code installed and running. If you do not, start with my 60-minute Claude Code guide, then come back here and pick a tool that solves a problem you actually have.
For how MCP servers and skills work under the hood, the official Anthropic Claude Code docs are the source of truth.
Claude Code tools FAQ.
What are the best tools to use with Claude Code?
It depends on the job. For context and memory I run token-optimizer, qmd, and Open Brain daily. For research and content there is last30days and the Qwoted backlinks skill. For building automations, n8n-mcp turns a plain-English brief into a working workflow. Superpowers changes how Claude approaches the work itself. The full list with my notes is above.
Do you actually use all of these tools and skills?
Yes. Everything on this page is wired into my own setup as an MCP server, an installed skill, or a CLI I run. I took off anything I was only testing. If it made the page, I use it.
Are these Claude Code tools and skills free?
Most of the open-source ones are free to install. You pay for the model usage, not the tool. free-claude-code can point at cheaper or local models, which is the whole reason it exists. A few, like Hyperlink, are commercial products.
How do I install a Claude Code skill from GitHub?
Some of these are plugins you add through a marketplace, like superpowers, the Obsidian skills, and Codex. Others drop into your .claude/skills folder or get added as an MCP server in your config. Each repo has its own README with the exact steps. Read it before you install, since a skill runs inside your sessions and you want to know what it touches.
What is a Claude Code MCP server?
MCP is the protocol that lets Claude Code talk to outside tools and data. qmd, Open Brain, and n8n-mcp all run as MCP servers, which is how Claude can search your local files, pull your saved context, or build a workflow without you pasting anything in. Anthropic's own docs cover the setup.
The Claude Code Blueprint.
Five interactive lessons. Install Claude Code, build your first automation, and deploy it live on the internet, all in under an hour. Free, no coding required.
Grab the Blueprint →