Claude With The Anthropic Api

All three models share Claude's core capabilities: text generation, coding, image analysis, and other tasks

Step 1:** User enters text in web app and clicks send

Step 1**: Install packages in Jupyter notebook using `%pip install anthropic python-dotenv

The Anthropic API and Claude do not store any messages sent to them

Prompt user for text input using the built-in `input()` function

System prompts customize the style and tone of Claude's responses

• **System prompts assign roles to Claude** - You can define a specific persona or role (e.g., "Python engineer") to guide the AI's behavior and response style

Claude tokenizes input text (breaks it into smaller chunks)

User messages sent to Claude can take 10-30 seconds to receive a response

Technique for generating **structured data** (JSON, Python code, bulleted lists) with no extra commentary

Generate three AWS CLI commands in a single response with no additional commentary or explanation

Prompt engineering**: techniques for writing/editing prompts to help Claude understand requests and respond as desired

No single standardized methodology for assembling evaluation workflows across the industry

Create a prompt to help users write AWS-specific code

Process each test case by merging it with a prompt, sending to Claude, and grading the output

A grader takes model output and returns an objective signal (commonly a number between 1-10, where 10 = high quality, 1 = low quality)

Code grader validates model output to ensure it returns only plain Python, JSON, or regular expressions without explanations

Goal: enhance model grader by providing context about what constitutes a good solution

Series of videos building on a single prompt iteratively

The **first line of your prompt is the most important** — use it to set the task clearly

Being specific means providing guidelines or steps to direct the model toward a particular output

Purpose**: XML tags clarify and organize content within prompts, especially when interpolating large amounts of data

One-shot prompting**: Providing a single example in your prompt

Goal: Improve an existing prompt to extract topics from scholarly article passages

Build a Jupyter notebook project to teach Claude how to set reminders for future dates

Tools allow Claude to retrieve extra information by executing Python functions automatically

JSON schema is a data validation specification—a set of rules for validating any JSON data

Include a `tools` keyword argument when calling `client.messages.create()

Access the tool use block from Claude's response via `response.content[1]

Goal is to wire up multiple different tools to Claude, not just one

Keep calling Claude in a loop until it stops requesting tool use

Tool Implementation Already Provided**: The code for `add_duration_to_date_time` and `set_reminder` functions was pre-built in the tools and schemas cell

Streaming delivers real-time API responses as a series of events

Claude has one built-in tool by default: the **Text Editor tool

Built-in tool that allows Claude to search the web for current or specialized information

A technique for answering specific questions about large documents by retrieving relevant content before generating responses

RAG workflow: source document → chunk into pieces → user query → retrieve relevant chunks → add to prompt context

Semantic search** is the process of finding text chunks related to a user's question

Step 1: Text Chunking** – Break source documents into separate pieces of text for processing

Implementing a complete RAG (Retrieval-Augmented Generation) flow using a vector database

Semantic search can miss relevant results in edge cases

Two search implementations (semantic/vector search and lexical/BM25 search) have nearly identical public APIs

Extended thinking allows Claude to reason about queries before generating final responses

Claude can process images included in user messages

Claude can read and process content directly from PDF files, not just images

Citations allow Claude to reference outside sources when generating answers

Claude performs extensive internal processing on input text before generating output (creating data structures and calculations)

Initial request to Claude is processed and the work is saved to a temporary cache

Prompt caching allows you to cache repeated content (system prompts, tool schemas, messages) to reduce token usage and costs

Upload files (PDFs, images, text files) to Claude and receive a unique **file ID

A communication layer that provides Claude with context and tools without requiring developers to write extensive custom code

Client provides communication between your server and an MCP server

Building a CLI-based chatbot to understand how MCP clients and servers work together

Building an MCP (Model Context Protocol) server to add functionality to an existing CLI chatbot

Activate your Python environment (check the readme for exact command)

The MCP client wraps a client session, which is the actual connection to the MCP server

Resources** allow MCP servers to expose data to clients

MCP client needs a function to request resources defined in the MCP server

Prompts are pre-built, well-tested message templates that MCP servers expose to clients

Call `self.session.list_prompts()` to retrieve all prompts defined in the MCP server

Tools**: Model controlled

CloudCode** is a terminal-based coding assistant built and deployed by Anthropic

Terminal-based coding assistant that runs in your terminal

Cloud Code is not just a code-writing tool—it's like having another engineer working alongside you on a project

Cloud Code has an embedded MCP (Model Context Protocol) client that allows connection to MCP servers

Workflows and agents** are strategies for handling user tasks that cannot be completed by Claude in a single request

Asking Claude to evaluate multiple material options in one prompt is inefficient

Prompt chaining** breaks down one large task into a series of smaller, sequential subtasks

Different topics require different video script styles and tones

Workflows** are best when you know the exact sequence of steps needed

Agents need to inspect their environment after (and sometimes before) taking actions to understand results

Predefined series of calls to Claude with a known sequence of steps

Anthropic offers **Haiku** (fast, smaller requests) and **Sonnet** (greater intelligence)