Amazon Digital Virtual Assistant (amzdva) [CLAUDE/AGENT SKILL]
A local Agent Skill that gives your AI assistant authenticated, programmatic read access to your Amazon Seller Central account. Drop the unzipped folder into Claude Code, Codex, Cursor, or any shell-capable agent. Say "install amzdva." The agent handles the rest - Node dependencies, Playwright browser, sign-in, all of it. About five minutes from zip to working.Live data, on demand - twenty-five commands covering every read-only SC surface that matters: amzdva kpi - pull your SALES, ORDERS, MESSAGES, BUYBOX, GLOBAL_ORDERS_WIDGET cards across all (brand, marketplace) pairs in one call. Filter by brand, by marketplace, or both. amzdva orders --us --sum-cancelled - total cancelled MFN + FBA orders across all marketplace brands, 30/60/90-day windows. The metric no other tool surfaces in one line. amzdva list-brands - every global account, every marketplace, with the merchant ID + marketplace ID you'll need for downstream tools. amzdva ranking --brand "your brand" - Brand Analytics → Search Query Performance, with rank, share of click, share of purchase per query. amzdva ba <dashboard> - universal Brand Analytics: query-performance, top-search-terms, brand-catalog-performance, demographics, market-basket, repeat-purchase, customer-journey, customer-loyalty. Talk to any BA dashboard Amazon exposes, with any filter combination. amzdva listing-quality - which listings have no offer, lost the buybox, no images, are out of stock, are suppressed. The kind of list you'd build by hand in Seller Central over an hour. amzdva acos - your Sponsored Products cross-program campaign report, sorted by spend / ACoS / ROAS / sales. amzdva report sales - the live Business Reports sales dashboard CSV. Always available, always current. amzdva fetch <path> - universal escape hatch. Any authenticated Seller Central endpoint, with auto CSRF + brand-injection. Use this when you discover an endpoint the other commands don't cover. amzdva capture --duration 60 - record every XHR the page makes for 60 seconds, so you can discover new endpoints without ever clicking around yourself. The agent drives, the user doesn't click - every command runs from your terminal, driven by the AI agent. You don't log in to Seller Central and click through menus. The agent fetches, parses, and reports. The only thing you do manually is the one-time sign-in (in the headed Chromium window the agent opens for you).Persistent session - after the first sign-in, the CLI saves your Seller Central cookies to ~/.amzdva/chrome-profile/. Every subsequent command reuses them. No daily re-auth unless Amazon forces a logout.Parallel-by-default, not "wait for the page" - most SC tools take the lazy path: open a page, click, wait for the React render, scrape. We take the data path. amzdva captures your session once, then fans out across every brand × marketplace × dashboard in parallel from Node. A full brand × 5-card KPI sweep that would take ~58 minutes on the page-scrape pattern finishes in ~25 seconds here. Same data, same correctness, ~140× faster. There's no await page.click() in the loop — just cookies + CSRF + Node fetch, the way it should have been built the first time.No external APIs - no RapidAPI, no OpenRouter, no Helium 10. The only thing amzdva needs is a working login to your Seller Central account. There is no LLM in the pipeline at all. The agent (Claude, GPT, etc.) is the reasoner; amzdva is the data layer. This separation is intentional - the data should be deterministic and reproducible, and the strategy should live in the agent you're already paying for.Retry-on-rate-limit, not "give up" - Amazon rate-limits SC aggressively, especially the inventory GraphQL endpoint. amzdva has retry-with-exponential-backoff baked into every fetch: a transient 429 or 503 backs off ~500ms then ~1s then ~2s and retries, honoring Retry-After headers. A sweep that hits 12 rate-limit errors recovers all of them in ~3 seconds instead of returning half a dataset.TRAIN your instance - The skill can LEARN new endpoints and data points. If the current commands don't capture the data you need, just tell the skill where it is, or ask it to discover, and it will find the data, save the endpoint, and deliver your custom information on demand.What makes it differentThe SC ecosystem has two kinds of tools. The cheap ones are dashboards: pretty graphs, but the data is what SC's homepage shows you - five cards per page, brand-by-brand, click-click-click. The expensive ones are agencies: $2k/month for a human to log in and read the same five cards, copy them into a spreadsheet, and email it to you. Neither one is fast, and neither one is something an AI agent can drive.amzdva is the third option: a deterministic CLI that an AI agent can call as a function. No scraping, no waiting for pages to render, no graphical dashboard to interpret. The agent asks, amzdva fetches, the agent reasons. The whole loop runs in seconds, not minutes, and the data is structured for downstream piping.The other thing that's different: the underlying architecture. Most SC tools use page-scrape patterns (open browser, navigate, wait for React render, read DOM). We don't. amzdva captures your session and CSRF once, then runs parallel Node fetches directly to SC's GraphQL BFFs. The browser tab is never navigated during the data-fetch loop. That's why a 116-brand sweep runs in 25 seconds instead of 58 minutes - it's the architectural difference, not a "faster scrape."What you do with it Daily VA cycle - at the start of each session, ask the agent: "Pull my sales for the past 7 days across all US brands, my cancelled-order count, and my top 3 listings by quality issues." One prompt, one response, three commands composed. Discovery - when you don't know the URL for the data you want, amzdva capture --duration 60 while you click around in your browser. The JSONL output lists every XHR. Pick the interesting ones. The agent uses amzdva fetch to call them. Build your own dashboards - pipe any amzdva <command> --json into jq, into a spreadsheet, into a Notion database. The output is structured, not screen-scraped. Health checks on a schedule - pair with amzdva watch for a long-running watcher that diffs KPIs every ~15 minutes and alerts you when something moves. Multi-account / multi-brand - every command takes --brand and --marketplace filters. Run across all 20 brands × 6 marketplaces in parallel. Why it's freeamzdva is the data layer. The strategic-analysis skills (keyword discovery, listing audit, PPC analyzer, SQP diagnostics) are paid because they take time to build, they call external APIs (RapidAPI for Amazon listing data, OpenRouter for the LLM labeling pass), and they produce decisions that pay for themselves in week one. amzdva by itself is a tool, not a strategy - but it's the foundation every other skill assumes you have. Free for you to use forever, MIT-licensed, fork it, modify it.What you needNode 20 or newer, a shell-capable AI agent (Claude Code, ChatGPT Codex, OpenCode, Google Antigravity, Cursor, Cline, Continue, Aider - the install runbook adapts to whichever you're using), and a Seller Central account. Windows, macOS, Linux all supported. The Playwright Chromium (~150 MB) downloads once on first install.What it doesn't do No write actions - read-only by design. This is a read tool. If you need an action, your agent recommends it; you apply it in Seller Central yourself. No search volume estimates, no rank tracking, no PPC bidding - different product. (For keywords, see the paid sister skill Amazon Keyword Discovery. For per-query funnel diagnosis, Amazon ASIN-Level SQP Diagnostic. For PPC search-term analysis, Amazon PPC Search Term Analyzer.) No listing audit (copy + images + reviews) - amzdva surfaces Amazon's listing-quality flags, but a real audit (vs. competitors, image compliance, review themes) is a separate skill. (Amazon Listing Auditor & Image Analyzer.) No Brand View / ASIN View SQP analysis - amzdva fetches the data; the strategic analysis (category gaps, conquest targets, funnel-stage diagnosis) is the sister skill. (Amazon Brand View SQP Analyzer / Amazon ASIN-Level SQP Diagnostic.) No direct Seller Central changes - we recommend; you apply. US-first - v0.2 is amazon.com. International SC IS SUPPORTED, but may need small tweaks (timezone, currency, marketplace IDs). One agent per browser session - two concurrent amzdva processes share the same cookies. Read-only is fine; don't run two amzdva login flows at once.
Get it → aiplus3.gumroad.com