The Claude Hub
Everything on Claude: projects, examples, MCP, skills, plugins, agents, and prompts. 35+ guides on every term, every build, and every comparison, organized by what you're trying to figure out.
Last updated: July 2026
Every guide I’ve written about Claude Code, organized by what you’re actually trying to figure out.
I’ve been building with Claude Code since early 2025. Over a year of daily use, a $1,600 bill I didn’t see coming, and 35 articles later.
This page is the thing I wish existed when I started. Every guide I’ve published about Claude and Claude Code, organized by what you’re actually trying to figure out.
If you’re looking for a specific Claude Code guide, scan the questions below. If you’re new and don’t know where to start, work down from the top, the order matters.
Jump to: What is Claude? · Install Claude · What to build · Better results · Cost control · MCP · Plugins, skills, modes · Production · Compare tools
What’s in this hub
Understanding Claude
What is Claude Code (non-developer) · What is Claude Cowork · Cowork complete guide · 17 real workflows · MCP vs Plugins vs Skills · Skills vs GPTs vs Gems · Plugins vs Custom GPT
Install & First Project
Install from scratch · Onboard Claude to your work
What to Build
Real problems, not app ideas · 15 projects in tier order · Validate your app idea in 70 min
Better Results
Best daily prompts · AI mentor setup · Reverse-engineer business models · AI research · Prompting principles · Subagents · Dynamic workflows
Cost Control
Why is my bill high (19 fixes) · Build without burning cash · Local models with Ollama
MCP & Connections
What is MCP · Apps/connectors/plugins explained · MCP server types · NotebookLM 9 workflows · NotebookLM recordings · Make money with AI · 16 best servers · Apify MCP (pay-per-result) · Multi-client setup · Custom research server · Gmail multi-account · Build custom MCPs · Perplexity research · Build from scratch
Plugins, Skills & Modes
Best plugins (11 tested) · Build a Claude skill · Fix broken skills · Remote modes (18 tasks) · Claude in Chrome · Migrate to redesign · Image generation · Routines (7 tasks) · Scheduling modes
Production
Vibe coding to production
Compare Tools
Claude vs 8 tools (3-stage audit) · Cursor complete guide · Cursor MCP setup · 4 automation levels
Claude Updates 2026
Apr redesign · Dispatch (5 modes) · Routines · Chrome · Scheduling modes · Dynamic Workflows
Hi, I’m Jenny 👋
I build AI systems and write about what happens when they run. What works, and what breaks.
This is the Claude Hub, other hubs you might also enjoy:
AI Agents · Vibe Coding · Shipped Products · Substack Growth
Understanding Claude
Three tools, one name. Claude is the AI you chat with. Claude Code is the terminal-based coding agent. Claude Cowork connects Claude to scheduled routines and background automation. Most guides conflate them. Most frustration comes from using the wrong one.
What is Claude Code? (The non-developer’s honest answer)
Most “what is Claude Code” explainers are aimed at developers who already know what a terminal is. This one isn’t. If you’ve used Claude chat and aren’t sure what Claude Code adds or whether it’s worth the switch, start here.
What is Claude Cowork?
Cowork vs Claude chat vs Claude Code, in plain terms. The confusion between these three is the most common reason builders end up using the wrong tool for the wrong job.
Claude Cowork for Professionals: The Complete Guide
The full practical playbook for using Cowork in actual professional workflows. Covers setup, how to write operating procedures Claude will follow, and the 5 job categories where Cowork consistently beats manual prompting.
Claude Cowork real workflows: 17 creators, 15 that work, 4 who walked away
Not a features list. 17 real workflows, with honest verdicts on which held up over time and which got abandoned. The 4 that failed tell you more than the 15 that worked.
MCP vs Plugins vs Skills in Claude Code: which one do you use when?
Three extension types that look similar from the outside and fail differently. This is the decision guide: what each one is for, what each one costs you, and the 3-question flow for choosing between them.
Claude Skills vs ChatGPT GPTs vs Gemini Gems: how they compare
Cross-platform comparison of AI assistant customization layers. If you’re deciding which platform’s extension model to invest time in, this is the side-by-side.
Claude Plugins vs Custom GPT vs Gemini Gem
Which AI extension model wins for real work tasks. Goes past the features comparison to actual output quality on the same task.
Install & First Project
Install takes under 30 minutes. The harder part is the onboarding pass that teaches Claude your actual files before you build anything. Most builders skip this and spend weeks fighting a Claude that doesn’t know their project.
How do I install Claude Code from scratch?
Written for builders who have never opened a terminal. Covers installation, first run, and the 3 configuration steps that change how Claude behaves from day one. No developer experience assumed.
How do I onboard Claude Code to my actual work?
The setup that adapts to your files, then maps its findings to features. This is not the same as “just open a folder.” This is a structured onboarding pass that ends with Claude knowing what you’re building and why.
What to Build
One sequence: local scripts and tools first, then APIs and databases, then systems that run themselves. Don’t shop for ideas. Work the path in order. The builders who stall are almost always skipping tiers.
How to use Claude Code for real problems, not just app ideas
The shift from “what app should I build” to “what daily problem can I solve today.” With 8 concrete problem patterns that apply whether you’re a writer, operator, or developer.
15 Claude Code projects in order — from first prompt to autonomous system
15 projects across 3 tiers: local and reversible, connected and stateful, autonomous and scheduled. The first prompt for each is included. The tier order matters more than the specific projects.
Research with Claude Code: Validate Your App Idea in 70 Minutes
16 sites researched in 70 minutes using Claude Code’s parallel agents. The full research stack, 9 copy-paste prompts for technical audits, business model analysis, and API testing, and a framework that applies to any niche. Built the week before a real launch to find out if the market was there.
How to Build a Personalized AI Research Agent
A 3-phase methodology that turns any AI coding tool into an autonomous researcher, demonstrated with pharmaceutical research and adaptable to any domain.
How to Build a Voice-First AI System That Thinks With You
Local Whisper, four processing modes, global hotkeys: voice input in every app on your computer, zero API costs.
Better Results
Output quality tracks context quality. The prompts I use daily are below. The patterns underneath them are what actually transfer.
How to Prompt AI to Build Any App
The 7-round conversation system that resolves architecture before you open Claude Code, with fill-in templates for every round.
LLMs don’t read, they parse: format your prompts so AI uses your words
Markdown, XML, and JSON aren’t interchangeable when you feed them to a model. The format you choose changes what the model actually attends to. This breaks down when each one wins, and why formatting your input is the cheapest quality lever you have.
What are the best Claude Code prompts to use daily?
15 prompts that save me 30 hours a week, grouped by workflow stage. Not a random collection. Each prompt is here because it solved a specific recurring problem and I’ve used it more than 50 times.
3 levels of the Claude workflow stack that close the knowledge work loop
Three layers of workflow, from single prompts to chained steps to a full loop that takes knowledge work from input to finished output. The map for where your current setup sits and what the next level looks like.
How I hired Hormozi, Welsh, and Koe as AI mentors in 30 minutes for free
The prompt setup that turns any creator into an on-demand mentor inside Claude. I use it when I want strategic feedback from a specific frame, not generic encouragement. Takes 30 minutes to set up, runs indefinitely.
This Claude Code prompt reverse-engineers any creator’s business model
Hormozi and Welsh broken down by the same prompt, with exact output included. The interesting part: they’re doing similar things in completely different ways, and the prompt surfaces both.
How to do research with AI: 3 questions walked through live
The sequence that turns a vague question into a grounded answer. Walked through on 3 real questions from start to final output. The process is the deliverable here, not the research results.
How do I prompt AI coding tools so they actually do what I want?
The principles that transfer across Claude Code, Cursor, and any AI coding environment. Not tool-specific tactics. The underlying mechanics that make prompts work regardless of which tool you’re in.
What are Claude Code subagents and how do I use them?
Claude Code can delegate work to mini-agents inside a single session. Each gets clean memory, handles one job, and reports back. The part most guides miss: the description field is what Claude reads when routing. Vague descriptions mean nothing gets delegated. This covers the built-in types, when to spawn them, and the 5 failure modes that look like success until they don’t.
Claude Code dynamic workflows: how 100 agents research, write, and build in one session
Dynamic workflows let Claude spawn parallel agents inside a single session, split the work, have them fact-check each other, and hand you a merged result. I asked for research. What came back had already argued with itself before I saw it. The second half of this article shows how to run the same 6-move pattern in plain Claude or ChatGPT if you don’t have Claude Code yet.
Cost Control
Token costs compound silently. The $1,600 bill arrived before I understood what was accumulating. Three articles on getting that under control.
Why is my Claude Code bill so high (and how do I fix it)?
The 19 changes that stopped my $1,600 month. They split into 3 categories: setup changes you configure once, session habits that prevent runaway context, and project patterns that stop Claude from re-reading files it already has. Most take under 10 minutes. The ones that took me the longest to figure out saved the most.
How do I build with AI without burning cash?
The broader playbook for building cheap from $0 to launch. Stack selection, when to use frontier models vs local ones, and the 3 spending traps builders hit at different stages.
How do I run local models in Claude Desktop?
Ollama setup that cuts API calls to zero for the workflows that don’t need frontier models. If you’re running repetitive tasks or drafting work, this pays back in the first week.
MCP & Connections
MCP turns Claude Code from a coding assistant into something connected to your actual work: Gmail, Substack, databases, any tool with an API. Start with what MCP is, then move to setup and building.
What is MCP and why does it change everything?
The concept guide: what MCP is, what it unlocks, and how to think about it before you install anything. This is where to start if you’ve heard the term and aren’t sure what it actually means in practice.
What are MCP apps, connectors, and plugins? The ecosystem explained
The MCP ecosystem has three layers that people constantly confuse. This maps them: apps, connectors, plugins, and which one to reach for depending on what you’re trying to connect.
Which MCP server type should I install, and how do I set it up?
stdio, HTTP, and OAuth MCP types explained with exact config file locations and fixes for the 4 connection errors you’ll hit. The type matters because they fail differently.
How do I set up MCP across Claude, ChatGPT, VS Code, and Cursor?
One setup walkthrough covers every major AI client. If you’re running Claude Code and Cursor side by side, this is how you connect the same servers to both without duplicating config work.
MCP Best Use Cases
How to use MCP to do real, practical work for you?
Best MCP servers for Claude Code — 16 ranked by real use
16 servers ranked by actual utility, not download counts. What’s worth connecting, what’s noise, and the 3 I use every single day.
The 4 best research MCPs for Claude (and the job each one owns)
Four research MCPs, each with the one job it does better than the rest. Which to reach for when you’re searching, scraping, reading papers, or querying your own sources, so you stop overlapping tools that do the same thing.
This MCP resolves 99% of your search problems. And it can pay you money.
The Apify MCP handles web search, scraping, and data extraction without per-search API subscriptions. It runs on a pay-per-result model: you pay only for what you use, and you can publish your own actors and get paid when others run them. I stumbled into $0.002 per run. The math compounds faster than it sounds.
I Handed My Substack Growth to Claude
Eight workflows, one MCP: digest your inbox, size a niche, study top creators, draft and schedule notes in your voice, and improve conversion.
Obsidian + Claude Code: build a second brain that actually answers you
Point Claude Code at your Obsidian vault and it stops being a note pile and starts being a source you can query, cross-link, and build on. The setup that turns years of scattered notes into a working knowledge base, with the wiring that keeps links and structure intact.
Claude + Canva MCP: on-brand designs on a free plan
The Canva MCP lets Claude generate real, editable Canva designs from your content, and it works on the free tier. What it can and can’t do without a paid plan, and the workflow for turning an article into on-brand graphics without leaving Claude.
Claude Code + NotebookLM: 9 workflows I run weekly
Claude Code handles reasoning and execution. NotebookLM holds and surfaces knowledge. These 9 workflows pair them for research, content, and analysis. Each one has a clear input, output, and time estimate.
How to query any recording without rewatching it (Claude + NotebookLM)
Turn any course, meeting, or video into a queryable source through the NotebookLM MCP. I use this to get answers from recordings I can’t afford to rewatch, and to extract specific moments across a library of calls.
How to make money with AI: from research to act in one loop
The Claude Code + NotebookLM loop turned into real revenue, end to end. The loop is real. The numbers are real. The workflow is documented.
How do I connect Gmail to Claude Code across multiple accounts?
One config covers all accounts. I run 4 Gmail addresses through Claude Code. This is the setup, and the 5 workflows it unlocked.
How do I build a custom MCP server for my own research?
A real example: 3,000+ Substack notes across 9 creators, pulled into a searchable research server. The three-pass methodology is reusable for any personal knowledge base you want to make queryable.
How do I build my own custom MCPs for Claude Code and Cursor?
The custom MCP stack that lives inside your subscription. When to build vs when to install something from the store, and what a working custom MCP looks like from start to finish.
How do I connect Perplexity to Claude Code for research?
Chrome MCP bridge setup, with a competitor research workflow built in 2 hours. The workflow scans 177 products across 5 platforms. Built the week before a product launch when I had no idea what the competitive landscape looked like.
How do I build an MCP server from scratch?
Complete step-by-step build guide, transport types, and a real working server. For when you need a connection that doesn’t exist in the store yet.
Plugins, Skills & Modes
Don’t install everything. Don’t migrate everything either. Tested verdicts below.
Which Claude Code plugins are worth installing?
11 plugins tested over 6 weeks on real work tasks. 4 made the cut. The framework for evaluating new ones is included, because the store grows faster than any list can keep up with.
How to build a Claude skill that works
Most skill tutorials show you how to create the file. They skip why skills fail to trigger, how Claude decides which skill to use, and what makes a skill description actually route correctly. This one covers all three. Built from debugging 20+ skills that weren’t running.
Why are my Claude skills not working, and how do I fix them?
An audit script that finds skills doing nothing in your stack. If you’ve built skills that Claude ignores, the problem is almost always in the description field, not the skill itself.
What can each Claude remote mode actually do?
18 real tasks across all 5 modes (Dispatch, Tasks, Computer, Chrome, code triggers). Exact prompts used, honest verdicts on what actually ran vs what got stuck.
What can Claude actually do inside Chrome?
5 tested use types from page reading to scheduled tasks. The setup fix that most guides miss is included. Start here before building any browser automation workflow.
How do I migrate my workflow to the new Claude Code redesign?
18 automation jobs sorted in 2 days, with a copy-paste Slack approval loop. If the redesign broke your existing setup, this is the migration path.
Which Claude Code image generation tool is worth it? (Nano Banana deep dive)
100+ images tested, real verdict on the tool that’s everywhere in the Claude Code ecosystem right now.
What are Claude Code routines and what should I actually schedule?
7 recurring tasks worth scheduling today. Each one has the tools it needs, the input it takes, the output it produces, and a starter prompt. The filter for what belongs in a routine: does the value come from it running consistently, without you initiating it?
Claude Code /loop: how to keep it running until the work is done
What /loop actually does, when to use it instead of a routine, and how to set an exit condition so it keeps working a task until it’s finished rather than stopping halfway.
Which Claude scheduling mode should I use?
4 modes, 4 failure patterns, one 5-question routing framework. Cloud Routines, Local Routines, Cowork Tasks, and /loop each break in different ways. The framework tells you which one fits the job before you build it.
Production
Going from “works on my machine” to deployed and maintainable is the 90% that isn’t writing the first draft.
How do I take a vibe-coded project to production without losing my mind?
The discipline that ships: planning before prompting, smoke testing before deploying, and the 3 conversations with Claude that catch what testing misses. The most-read article I’ve written. Start here.
The Software Engineering Practices Every AI Builder Actually Needs
The SE fundamentals that prevent the disasters AI tools can’t catch on their own. Covers the critical few practices that solve 80% of build-to-launch problems — not a textbook survey, a filtered list for people shipping with AI assistance. 133 likes, one of the most-saved articles in the catalog.
Compare Tools
Claude Code isn’t the only option. What you’re building determines the right tool.
How does Claude Code compare to Cursor, n8n, and 6 other AI coding tools?
A 3-stage audit on the same 12-section website spec, run across 8 tools. Not a features comparison. Actual output, actual time, actual cost. Decision flowchart by use case at the end.
How do I get the most out of Cursor end to end?
Full Cursor playbook: commands, MCP, and modes. The wiring patterns transfer directly to Claude Code if you’re running both.
Cursor 2.0 Is Rewriting the Future of AI Coding
A hands-on deep dive into the tool that’s blurring the line between coding, creation, and collaboration.
How do I set up Cursor with MCP?
Cursor-specific MCP and rules setup. The connection patterns work for Claude Code too.
4 levels of AI automation: when Claude, n8n, and OpenClaw each win
A decision framework for choosing between Claude-native, n8n, and OpenClaw by task type and complexity. Most builders jump to the highest level and overshoot. This maps out why that’s usually the wrong call.
Claude Updates 2026
Claude changed materially in 2026. The old “best prompts” framing is too small for what the product became. What shifted: the product unbundled into surfaces (Chat, Cowork, Code, Design), each with different permissions, cost structures, and use cases. Choosing the wrong surface is now the most common reason workflows fail.
Claude Cowork: What It Is and How It’s Different from Claude Code
January 2026. Claude unbundled. Cowork arrived as a separate surface — scheduled context, persistent memory, operating procedure support. If you’ve been treating Claude as a single product, this is the article that maps when each surface applies. The confusion between Cowork, Code, and Chat drove most of the “why isn’t this working” questions in the first half of 2026.
MCP vs Plugins vs Skills: The 2026 Extension Model, Explained
April 2026. Three extension types that look similar and fail differently. The extension model didn’t change — but the product grew until the distinctions mattered enough to document. This is the decision guide: what each one is for, what each one costs, and the 3-question routing flow.
Custom MCPs Move into the Subscription
April 2026. Custom MCP servers went from side project to first-class capability inside the Claude Code subscription. Build your own or connect ones from the store — this covers both, with a real working example. The shift that made Claude Code meaningfully more powerful than a standard AI coding tool.
Image Generation Lands in Claude Code (Nano Banana Deep Dive)
April 2026. Native image generation for Claude Code via the Nano Banana plugin. 100+ images tested, real verdict on output quality and where it breaks. The plugin spread faster through the Claude Code ecosystem in Q2 2026 than anything else in the store — worth knowing what it actually produces before committing.
Claude Cowork for Professionals: The Full Playbook
May 2026. Cowork had been live for four months. This is the guide that arrived once people had used it long enough to know what works: how to write operating procedures Claude will follow, the 5 job categories where Cowork consistently outperforms manual prompting, and where it fails in ways Claude Code doesn’t.
Claude Code April 2026 Redesign: How It Changed vs Cursor and OpenClaw
The April redesign introduced Routines and reorganized how Claude Code sits relative to Cowork and third-party tools. Covered here: what actually changed, how it shifts the Claude vs Cursor comparison, and a migration map for 18 existing automation jobs.
Claude Dispatch: 5 Remote Modes, 18 Tasks Tested
Dispatch, Scheduled Tasks, Computer Use, Chrome, and code triggers — all 5 modes with exact prompts and honest verdicts on what ran and what got stuck. The 80% the demos don’t show.
Claude Routines: What Launched and What’s Worth Scheduling
Routines landed as the primary recurring automation primitive. 7 use cases worth scheduling, with tools, inputs, outputs, and starter prompts for each.
Claude in Chrome: What the New Browser Mode Actually Does
Chrome integration shipped with less coverage than the other surfaces. Tested across 5 use types — from page reading to scheduled browser tasks — with the setup fix most guides skip.
Claude Scheduling in 2026: 4 Modes, One Decision Framework
Cowork Tasks, Local Routines, Cloud Routines, and /loop arrived within months of each other with no clear product map. This is that map: what each one does, where each one breaks, and the 5-question routing test.
Dynamic Workflows: 100 Agents in One Session
June 2026. Claude Code can now spin up 100 parallel agents inside a single session — each gets clean memory, handles one job, fact-checks the others, and hands back a merged result. What the feature does and how to run the same pattern in any AI tool.
FAQ
What is Claude Code, really?
Claude Code is Anthropic’s terminal-based coding agent. You run it from the command line, point it at a codebase or project folder, and it reads files, writes code, runs tests, and makes commits. Unlike Claude chat, it has direct access to your filesystem. Unlike GitHub Copilot, it works on multi-step tasks across many files at once. The closest mental model: a developer who can read your entire project in one pass and then act on it.
Do I need to be a developer to use Claude Code?
No, but you need to be comfortable with a terminal. If you can open Terminal on a Mac and run one command, you have enough. The beginner’s install guide was written for non-developers and takes 30 minutes. What Claude Code does well for non-developers: writing scripts, building automation tools, and creating things that would otherwise require hiring someone. What it does poorly: real-time IDE assistance while you write code yourself. That’s Cursor’s job.
How is Claude Code different from Cursor?
Cursor is an IDE you write code inside. Claude Code is an agent you give tasks to. Use Cursor when you’re writing code and want inline suggestions. Use Claude Code when you want to hand off a multi-step task and come back to a result. Most builders who use both end up with Cursor for fast edits and Claude Code for anything that requires planning, coordination across files, or running processes. The full comparison is here.
What does Claude Code actually cost?
You pay per token: roughly $3 per million input tokens and $15 per million output tokens on Claude Sonnet. A typical session runs 20,000–80,000 tokens. A complex session with file reading, iteration, and testing can hit 300,000+. My $1,600 month came from not knowing what was compounding. The token optimization article covers the 19 changes that brought my bill down 40-70%.
What is CLAUDE.md?
CLAUDE.md is a plain text file you put in the root of your project. Claude Code reads it at the start of every session. It’s where you write your project’s context, rules, and preferences so Claude doesn’t have to ask or infer. Think of it as a standing briefing: what this project is, how it’s structured, what Claude should and shouldn’t do. A well-written CLAUDE.md eliminates most of the repetitive setup prompting that slows sessions down.
How is this different from Anthropic’s documentation?
Anthropic’s docs explain what Claude Code can do. This hub explains what to actually build, what breaks first, and what costs you money. Every article here came from a real failure, a real bill, or a workflow I had to build from scratch. The docs are the product map. This is the field map. Different job.
How often is this hub updated?
Every time I publish a new Claude-related guide, it gets added here. The “Last updated” date at the top reflects the most recent addition. The hub has grown from 3 articles to 25+ since early 2025. If something you’re looking for isn’t here yet, it’s probably in progress.
The path through this hub
What is Claude → Install → first project → prompts and subagents → cost control → MCP → plugins, skills, and modes → production → compare alternatives. Same sequence whether you stay on Claude Code or switch between tools.
Browse all articles by topic: Claude Code · Prompts and Templates