Claude

Sign up and understand the interface

Go to claude.ai → Sign up with Google or email. Free plan: Claude Sonnet with daily message limits. Pro ($20/mo): Claude Opus 4.5, 5x more usage, Projects with file storage, and Priority access.

The interface: left sidebar has conversation history and Projects. Main area is the chat. Paperclip icon uploads files. The model name shown top-center tells you which model is active.

First test: paste any 2-3 paragraph document and ask Summarize this in 3 bullet points and identify the single most important claim. Notice how Claude reads the full text, not just skims it.

Analyze your first document from SEC EDGAR or anywhere

This is Claude's strongest use case. Download any real document — a company annual report (10-K), a research paper, or a contract. Upload it via the paperclip icon.

Use this analysis prompt:
I have uploaded [Document Name]. Analyze it and give me: 1) A 3-sentence executive summary anyone can understand. 2) The 3 most important findings or decisions. 3) Anything that seems risky, unusual, or worth investigating further — be specific. 4) The 3 questions this document raises but does not answer. Use specific page or section references.

Claude reads the entire 100+ page document — including footnotes — and synthesizes it in under 60 seconds. This alone replaces hours of manual reading for analysts, lawyers, and researchers.

Create a Project with custom instructions and upload knowledge files

Click Projects → New Project. Name it clearly (e.g., Work — Legal, Client A, Coding — Python).

Write your instructions in the text field. Use this structure:
About me: I am a [role] at [company type]. Context: [describe what you do]. My preferences: [output format, writing style, level of detail]. Always: [specific rules Claude must follow]. Never: [what Claude must avoid]. When I ask for [task type], always include [specific thing].

Then upload your knowledge files: click Add content → upload PDFs, code files, style guides, research papers, or documents. Every conversation in this Project uses these files as context. Claude can answer questions about any uploaded content across all files simultaneously.

Write production-quality code from a description

Paste code and use this specific prompt structure to get professional results:
Review this [language] code as a senior engineer focused on correctness, performance, and maintainability. For each issue: 1) Category: [BUG / SECURITY / PERFORMANCE / READABILITY]. 2) Explain the problem and why it matters. 3) Show the fixed version side-by-side. End with an overall rating 1-10 and explanation.

For generating new code:
Write a [language] function that [exact description]. Requirements: [list]. Edge cases to handle: [list]. Include: type annotations, JSDoc/docstring, error handling for [specific failures], and 3 unit test examples. Use [library/pattern] over alternatives.

Build a multi-document research synthesis

This is Claude's most unique capability. Upload 4-8 documents on the same topic and ask questions that would take hours to answer manually.

Example: upload 6 competitor pricing pages, then ask:
Across all these competitor documents: 1) What pricing models are they using (per-seat, usage-based, flat)? 2) What features does every competitor include in their base plan? 3) What features do they charge extra for? 4) What are they NOT offering that might be a market gap? 5) How does each position their value proposition? Create a comparison table for the first 3 points.

Claude synthesizes across all documents simultaneously. No copy-pasting between chats, no manual cross-referencing.

Install a Custom Skill for your most repetitive task

Create a file named SKILL.md in a new folder:

---
name: Code Review
description: Structured code review following our standards. Invoke when asked to review code.
---

## Review Process
1. Bugs: logic errors, null pointer risks, race conditions
2. Security: injection, auth gaps, data exposure
3. Performance: N+1 queries, inefficient loops, blocking calls
4. Readability: unclear naming, missing types, no docstrings

## Output Format
For each issue: [SEVERITY: Critical/High/Medium/Low] description fix
End: Overall score /10 and the single most important change.

Connect the Claude API for your first integration

Install the SDK:

npm install @anthropic-ai/sdk  # Node.js
pip install anthropic          # Python

import Anthropic from '@anthropic-ai/sdk';

const client = new Anthropic();
// ANTHROPIC_API_KEY env variable auto-detected

const response = await client.messages.create({
  model: 'claude-opus-4-6',
  max_tokens: 1024,
  system: 'You are a helpful assistant.',
  messages: [{ role: 'user', content: 'Hello!' }]
});

console.log(response.content[0].text);

Build your first long-form content piece with Claude

Claude produces the most factually consistent long-form content of any AI. Use this workflow for a 2,000-word article:

Step 1 — Brief Claude:
I am writing a 2,000-word article for [audience] about [topic]. Core argument: [one sentence]. The main question the reader wants answered: [question]. Tone: [tone]. Before writing, ask me any questions that would make the article significantly better.

Step 2 — Approve the outline before writing begins.

Step 3 — Section by section:
Write section 2 from the approved outline. Include [specific example]. End with a transition to section 3. Show your reasoning.

Section-by-section writing consistently produces better output than asking for the full article at once.