Forem: Andre Moreira

Using Claude for Chrome

Andre Moreira — Tue, 30 Sep 2025 17:37:00 +0000

Overview

Claude for Chrome is a browser extension that brings AI assistance directly into your web
browsing experience. This guide demonstrates three practical examples of how Claude for
Chrome can enhance everyday productivity tasks, from email writing to document editing
and formatting.
Getting Started
Prerequisites

● A Claude account (Pro accounts received early access)
● Chrome browser
● Claude for Chrome extension installed
Accessing Claude for Chrome
Click the extension icon in your Chrome toolbar (where all extensions are located). This
opens a panel where you can interact with Claude while working in your browser tabs.
Key Feature: Saved Prompts
Claude for Chrome allows you to save frequently used prompts for quick reuse. This is
particularly useful for repetitive tasks like email correction or formatting operations. Test
different prompts over time and save the ones that work best for your workflow.
Practical Example 1: Email Improvement

Use Case: Quickly drafting an email and refining it before sending.

Workflow:

Open your email and click "Reply"
Draft a quick, rough message
Open the Claude for Chrome extension
Use a saved prompt (e.g., "Read and memorize the highlighted text and improve it")
Highlight the draft text
Let Claude refine and improve the message

Result: Claude rewrites the email with improvements to tone, clarity, and professionalism,
and even explains what changes it made.

Practical Example 2: Document Editing with Multiple

Instructions

Use Case: Editing a client proposal with several requested changes throughout the
document.
Workflow:

Open your document (e.g., a proposal in a web editor)
Place instructions directly in the text where changes are needed
Highlight the document and activate Claude
Tell Claude to "make the edits directly" How Claude Works: ● Reads through the entire document first to find all instructions ● Works systematically through each instruction ● Deletes instruction text as it completes each task ● Modifies the content according to your specifications Example Instructions: ● Tighten a specific section ● Create a summary table with project phases and timelines ● Reformat or restructure content Advanced Capability: Claude can create tables directly in the document, populating them cell by cell based on information in the text. It even double-checks its work at the end to ensure all instructions were completed.

Practical Example 3: Converting Bullet Points to Tables

Use Case: Transforming unstructured bullet points into organized tables.
Workflow:

Prepare or reuse a saved prompt that defines table structure (column names, format)
Customize the prompt for your current needs (e.g., change "Attribute" to "Topic")
Highlight the bullet point text
Run the prompt

Result: Claude creates a properly formatted table and systematically fills it with content from
the bullet points, matching each point to the appropriate column.

Key Benefits

Autonomous Operation
Claude can work on tasks while you do something else. When finished, it sends a
notification so you can review the results.

Agentic Behavior
Claude demonstrates sophisticated problem-solving:
● Reads and understands complex, multi-step instructions
● Plans its approach before executing
● Works systematically through tasks
● Self-checks to verify completion
● Adapts to document structure and formatting requirements
Real-World Integration
Particularly powerful for users who work extensively in Chrome, integrating AI assistance
directly into everyday workflows without switching contexts.

Technical Impressiveness
The system's ability to:
● Locate correct UI elements (buttons, cells, formatting options)
● Navigate complex document structures
● Create and populate tables in web-based editors
● Maintain context across multiple instructions
represents significant technical achievement in browser automation and AI agency.

Tips for Success

Use saved prompts for recurring tasks - refine them over time
Highlight text to ensure Claude knows exactly what to work with
Be specific in your instructions, especially for formatting tasks
Let Claude work autonomously on multi-step tasks while you focus elsewhere
Review Claude's explanations of what it changed to learn and improve future prompts

Current Status
As of the video recording, Claude for Chrome was in early access for Pro plan subscribers.
The feature is expected to roll out to a wider audience due to its significant productivity
benefits.
Workflow Summary

Install Claude for Chrome extension
Create and save useful prompts for common tasks
Work on your content in Chrome (emails, documents, web apps)
Activate Claude and use saved or custom prompts
Highlight relevant text to provide context
Let Claude execute the task autonomously
Review results and iterate as needed This guide accompanies the video demonstration of Claude for Chrome for everyday productivity applications.

Using Claude Code for non-coding tasks

Andre Moreira — Thu, 25 Sep 2025 20:45:10 +0000

I am a big fan of Claude Code. I like it so much that I also use it beyond traditional coding, transforming it into my personal deep researcher. The approach requires some initial setup, but it is worth it:

Claude account (Pro/Max recommended)
Claude Code installation
Visual Studio Code

I prepared a video where I show a couple of examples:

Two examples how I used it

EU Tender Analysis: Automated review of European Commission tender documents, matching company capabilities against tender requirements with contextual analysis of team expertise.
(here is another way to scan for tenders, in this case using n8n).
Market Research: Competitive landscape analysis for logistics equipment marketplace, including automated web research and comprehensive competitor dossier generation.

The key hack here is that this setup gives you a powerful deep-research assistant that works directly with your local files, while keeping Claude’s memory neatly compartmentalized. Each folder is its own project, so nothing overlaps. In practice, this means you can have specialized junior assistants (one per folder) handling time-intensive workflows in the background, and you simply review the results once they are finished.

There is an excellent course for anyone who wants to go deeper into Claude Code.

I hope this is helpful!

Using Agents for Business Development

Andre Moreira — Fri, 29 Aug 2025 06:43:59 +0000

This is a submission for the AI Agents Challenge powered by n8n and Bright Data

What I Built: Tradeshow Scanner

An intelligent lead qualification workflow powered by n8n AI Agent, Bright Data, and GPT-5-mini

Trade shows are goldmines for business development, but manually researching hundreds of exhibitors is time-consuming and inefficient. This workflow automates the entire prospecting pipeline for trade show attendees.

What it does

The workflow intelligently scans and qualifies potential clients from any trade show exhibitor list:

Automated Data Extraction: Uses Bright Data's web scraper to fetch the full exhibitor list from the trade show website, then extracts company names, descriptions, booth numbers, and website URLs using HTML parsing
Smart Batching: Processes companies in configurable batches (currently set to a small number for testing, easily adjustable) to manage API costs during development
AI-Powered Research & Qualification: Each company goes through an AI Agent that:
- Conducts web research using SerpAPI and targeted website scraping
- Evaluates company size (small/medium/large based on employee count)
- Assesses their public stance on data, AI, and automation initiatives
- Scores each company (low/medium/high) as a potential client based on your specific criteria
- Provides reasoning for each score to help prioritize follow-ups
Automated Documentation: Results are automatically appended to a Google Sheet with structured data including company details, qualification scores, and AI reasoning

The Intelligence Layer

The AI Agent is configured with detailed context about my company (lyfX.ai). It specifically looks for companies that match our profile and an "ideal" client profile: small-to-medium businesses with a modern, future-oriented approach who value working with specialized, high-caliber teams.

This creates a pre-qualified, prioritized list of booth visits and follow-up targets, transforming what would be hours of manual research into an automated, intelligent filtering system.

Demo

Video

https://www.loom.com/share/978e4d64c11b405aa85c8d72aca8b6a1?sid=0be6cc43-ea17-461a-9f1d-4e563cd8edac

n8n Workflow

https://gist.github.com/andremoreira73/fe972f825a5f41e36c8df5ad7cb1f2bb

Note that the workflow has been sanitized so anyone can adpat it to their own use case. I left the prompts intact, as they may serve as inspiration for others.

Technical Implementation

I designed this workflow with simplicity in mind - keeping the implementation clean while maintaining effectiveness. The HTML parsing and code transformations are minimal, just enough to structure the data and make it digestible for the AI agent. (The JavaScript in the code node was kindly prepared by my good friend Claude 😊)

Model Choice: I am using GPT-5-mini, which I have tested across other setups. What I particularly appreciate for agent workflows is OpenAI's consistent balance between reasoning capability and reliable structured output formatting. After nearly 2 years with the OpenAI API, there is also the practical advantage of familiarity and development speed.

Memory Architecture: This initial version runs stateless - each company evaluation is independent. For a future version, I am planning to implement memory for more sophisticated research patterns and cross-referencing between companies.

Tool Stack: Kept it focused with just two tools:

SerpAPI for web search - reliable and fast for internet-wide search
Bright Data for targeted scraping - handles sites that the agent may want to review

System Instructions: My favorite part! This is where the magic happens - carefully prepared a system prompt with instructions that encode our ideal client profile and evaluation criteria, ensuring consistent and relevant qualification scoring.

The beauty of this approach is its flexibility - swap the trade show URL, adjust the target profile in the system instructions, and you have a reusable lead qualification system for any event.

< Background >
The user provides a list of companies that will be participating in a trade show. The list contain the company's name, description, stand and website.
I will be representing my company (lyfX.ai) at the trade show.

About lyfX.ai: We help organizations integrate Data and AI solutions that address their business challenges and deliver measurable results. By merging deep business acumen, domain knowledge, data science, and modern AI, we deliver strategies that drive growth, enhance efficiency, and increase profitability.  We have deep knowhow in chemicals, process engineering, fertilizers and adjacent industries. 

We are a small team with very high caliber people: PhDs, one is a full professor at a reputable university in Germany, the other had a long successful career in the chemicals and fertilizer industries. Our team members hold different certifications: e.g. Google Associate Cloud Engineer, Google Professional ML Engineer, among others.

We are proficient in python, LangGraph, n8n, django.
</ Background >

< Objective >
Find out from the list of companies, which ones we should approach as potential new clients.
</ Objective >

< Target client profile >
- small to medium size company
- expresses publicly a high interest in data, AI, automation
- modern and future oriented
- not afraid of working with a small, high caliber team
</ Target client profile >

< Instructions > 
For each company in the list, do the following:
1) Find out what they do 
2) Is this a small (less than 50 employees), medium (between 10 and 1000 employees) or large company (over 1000 employees)? 
3) Based on what you found out, rate how this company scores as a target client profile: low, medium or high?

Be thorough and stick to the facts. 
If you don't know something or you don't find something, leave it blank.


</ Instructions >  

< Tools >
SerpApi: Google as search engine for internet search
Scraper: Scrape a specific URL as needed
< /Tools >

< Answer format > 
JSON format with the elements:
- name 
- description 
- stand 
- website 
- score as a target client profile
- reasoning for the score 
</ Answer format >

Bright Data Verified Node

I used Bright Data at two critical points in the workflow:

Initial Data Extraction: Bright Data's web scraper handles the complete exhibitor list extraction from trade show websites, reliably parsing pages (JavaScript-heavy pages, dynamic content, etc).
As an agent tool for its research: Bright Data enables the AI to perform targeted scraping of individual company websites on-demand. This allows the agent to gather real-time public information about company size, technology focus, and initiatives directly from source.

The integration was easy. Bright Data's n8n nodes required minimal configuration and handled complex sites that typically require browser automation.

The reliability meant zero manual intervention even when processing hundreds of exhibitors, making it ideal for production use cases where consistency matters.

Journey

This project was an excellent opportunity to explore n8n and Bright Data in depth. While I am invested in LangGraph, I have discovered that n8n excels at rapid prototyping in ways LangGraph does not. My usual workflow development involves numerous Jupyter notebooks for experimenting with graphs and agents, but n8n has opened up a new avenue: I can now create functional prototypes in hours rather than days, then either productionize them in LangGraph or keep them running in n8n directly.

Using Bright Data was a great experience. Previously, I had been using other scraping solutions, but Bright Data's approach is really elegant and robust. The AI-assisted collector builder particularly impressed me (though this workflow ultimately didn't require it). What stands out is the platform's versatility. The tool has definitely earned a permanent place in my tech stack.

Next steps

The combination of n8n's visual workflow building and Bright Data's reliable data extraction creates a powerful environment that bridges the gap between "quick experiment" and "production-ready solution."

This sweet spot is exactly what I have been looking for in my automation projects. I already created a few more business development workflows and I am looking forward to sharing them over time!