ebrahimelbagory/claude-fba

claude-fba is a toolkit for Amazon sellers who want to run their store operations from the terminal. It pairs Claude Code, an AI assistant that works in the command line, with Amazon's official Selling Partner API and optionally Helium 10, a paid market research service. The author built it while running his own real store and reports concrete results from day one: a full audit of 19 products in roughly 25 API calls, 437 review requests sent through Amazon's official system, and a mislabeled listing fixed that was converting at 2.7% while similar products hit 10 to 14%. The project is organized around five skills, each handling one part of an audit, fix, and measure loop. The store audit skill checks every listing for quality, keywords, rank, reviews, and stock, then compares against the previous audit to show what changed. Review automation sends Amazon's official review request for every eligible order and keeps doing it daily. Listing guard lets you edit live listing text safely, requiring your approval on exact wording and saving old versions for one-command undo. Impact tracker captures a baseline before changes and checks traffic, conversion, and revenue weekly afterward. Keyword research shows what competitors rank for that you do not, sorted by how affordable each keyword is to win. Setup involves cloning the repository, copying the skills into a .claude directory, filling in your API credentials, and verifying authentication with a Python script. You need Claude Code, Python 3.9 or higher, and a self-authorized SP-API app that takes about 30 minutes to set up. Helium 10 is optional and only needed for keyword and competitor data. The Python scripts use only the standard library, so there are no pip installs. Safety is a central concern since the tool touches a live store with real money. Everything is read-only by default. Any change visible to buyers goes through a draft, your approval, and execution, with backups before and verification after. All actions use official APIs, nothing is scraped, and the author says nothing violates Amazon's terms of service.