Understanding Claude

Claude is an AI assistant made by Anthropic. Unlike a search engine, it doesn't just look things up. It reads what you give it, thinks about it, and produces original output. It can write, analyse, summarise, plan, code, and reason across a huge range of tasks - and when it needs current information, it can search the web too.

Think of it as a capable colleague who is always available, never forgets what you've told it in the current conversation, and improves every time you give it feedback.

The quality of Claude's output is directly tied to the quality of your prompt. A vague prompt gets a vague response. A specific prompt, with audience, format, tone, and length, gets something you can actually use.

A token is a small chunk of text, roughly three-quarters of a word in English. When you send Claude a message, it breaks your text into tokens, processes them, and returns tokens as a response. Everything counts: your input, any files you attach, and Claude's reply.

You don't need to count tokens manually, but understanding they exist explains two things: why conversations have limits and why a concise, well-structured prompt often outperforms a rambling one. One specific prompt that nails it first time uses fewer tokens than five vague attempts.

The context window is the total amount of text Claude can hold at once in a single conversation. It includes everything: your messages, any files or documents you share, previous messages in the conversation, and Claude's own replies. When it fills up, the oldest content drops off.

A model is the engine behind Claude. Larger models handle harder tasks but are slower and use more of your allowance. For the vast majority of your day-to-day work, Sonnet is the right choice. Only reach for Opus when Sonnet's output isn't quite hitting the mark.

By default, every new conversation starts from zero. Claude also has a Memory feature that lets it remember key details - like your preferences, projects, and working style - across conversations. Your admin controls whether Memory is enabled for the team.

Keep going in the same conversation when you're refining one piece of work (e.g. iterating on a report). Start fresh when you're moving to a genuinely new task. If you need to bring context forward and Memory isn't covering it, paste in the key points.

Claude isn't limited to typing responses in a chat window. It can use tools - built-in capabilities that let it take action. It can search the web for live information, run code to crunch numbers, and create actual files (Word docs, spreadsheets, PDFs, presentations) you can download and share. When you ask Claude to "write a report," it produces a real document, not just text in the chat.

Tools are what turn Claude from a writing assistant into a working assistant. And as you'll see next, MCP lets you add even more tools by connecting Claude to external apps like Slack, Jira, and your databases.

The Model Context Protocol (MCP) is a universal standard that connects Claude to external apps: Slack, Jira, Gmail, Figma, Google Drive, databases, and more. Each connection gives Claude tools — specific actions it can perform inside that app, like "read a Slack channel," "create a Jira ticket," or "send an email." Before MCP, every integration was custom-built. MCP standardises the connection so any compatible app can plug in instantly.

You'll hear these connections called connectors. When you enable the Gmail connector, for example, Claude gains tools to search, read, and draft emails. Each connector is powered by MCP under the hood.

Three concepts work together here. You already know connectors — they link Claude to external apps. A skill is a saved set of instructions that tells Claude how to do a job: what data to pull, how to process it, and what format to deliver. Instead of explaining "write the report in this format, with this structure, in this tone" every time, you save it as a skill and Claude follows it automatically.

A plugin bundles everything together - skills, connectors, and other components - into something you can install in one click. Think of it as an "app" for Claude: it adds a whole workflow rather than just a single instruction.