Claude Cowork for Amazon Sellers: The 2026 Guide to Workflows, Plugins, and MCP Connectors

Claude Cowork is the right place to start if you're an Amazon seller who wants AI to do real operational work — not just answer questions, but actually move files, read your Seller Central reports, draft your Plans of Action, build your weekly business briefing, and run on a schedule while you sleep.

It is also, for any seller serious about Amazon-specific work, a starting point and not a destination.

This guide is the honest map. You'll get the three operating modes Cowork gives you (Dispatch, Scheduled Tasks, and Live Artifacts), the plugins and skills worth installing for Amazon work, the MCP servers that connect Cowork to Seller Central, the four workflows that hold up best in practice, and a clear-eyed look at where general-purpose Cowork stops being enough — and where a purpose-built Amazon platform takes over.

If you've spent any time experimenting with Cowork, you already know the first 80% feels like magic. This article is about both halves of the curve.

What Claude Cowork Actually Is

Claude Cowork is the agentic mode of the Claude desktop app — the same architecture that powers Claude Code, but wrapped in a UI built for non-developers. It launched on Windows on February 10, 2026 with full feature parity with macOS, which matters because the Amazon seller universe runs heavily on Windows.

The thing to internalize is that Cowork is not a chatbot. It has direct file access on your computer, a sandboxed Linux shell for running code, the ability to connect to external services through MCP connectors, a skills system for repeatable workflows, a plugin system for bundling those skills together, and a scheduler for running tasks automatically. For an Amazon seller, that combination is closer to a junior operations analyst with API access than to ChatGPT.

There are three modes you'll use.

Dispatch is the one-off mode. "Read this 47-page Amazon Plan of Action rejection letter, identify every reason Amazon listed, and draft a revised response that addresses each one with specific evidence from my account." Cowork goes to work, reads the file, drafts the response, and hands it back. You can fire a Dispatch and walk away — the work continues in the background, which is the part most chatbot users miss the first time they try it.

Scheduled Tasks are the recurring mode. "Every Sunday afternoon, generate a weekly content calendar for SellerForge. Every Monday morning at 7 a.m., pull last week's Seller Central business report and email me a one-page summary with anomalies highlighted." Scheduled tasks fire automatically as long as your computer is awake and the Claude desktop app is open. That last constraint matters and we'll come back to it.

Live Artifacts are persistent HTML pages that pull fresh data each time you open them. Build one once — a "my Amazon dashboard" page that re-queries your connectors every time you open it — and it lives on in your Cowork session, updating without further prompting. This is the closest Cowork comes to a dashboard product, and it's underused.

Layered on top of those three modes are four building blocks: skills, plugins, MCP connectors, and projects.

Skills are small markdown files that teach Claude how to do one specific thing. Anthropic ships a set of core skills (docx for Word documents, xlsx for spreadsheets, pptx for slide decks, pdf for PDFs, schedule for setting up scheduled tasks). You can build your own and Claude triggers them automatically when relevant context appears in a conversation.

Plugins are bundles of skills, connectors, slash commands, and sub-agents grouped around a specific job function. Install a plugin once and Claude gains a whole toolkit at once — rather than configuring each piece individually. Anthropic ships a public plugin marketplace; enterprises can run private marketplaces with admin-controlled provisioning.

MCP connectors are how Cowork talks to external systems. MCP (Model Context Protocol) is the open standard Anthropic introduced in late 2024; by 2026 the connector library covers Gmail, Google Drive, Google Calendar, Slack, Notion, Salesforce, DocuSign, Apollo, Clay, Outreach, FactSet, WordPress, and a long tail of third-party servers including several built specifically for Amazon Seller Central.

Projects are persistent workspaces with their own files, links, instructions, and memory. Shipped in the March 2026 update. For a seller, the right pattern is one project per brand or per account.

That's the toolbox. Now let's talk about what you actually build with it.

What to Build First: Four Amazon Workflows That Hold Up

The mistake most new Cowork users make is starting with the most ambitious workflow they can imagine. Build something useful and small first. Here are the four I'd recommend, in order.

1. The "Plan of Action" Dispatch

Amazon's account suspension and listing suppression process is a paperwork problem disguised as a compliance problem. A POA that gets reinstated has four sections — root cause, immediate corrective actions, long-term preventive actions, and supporting evidence — and Seller Performance rejects appeals that miss any one of them.

Cowork handles this well as a Dispatch:

1. 1Drop your Account Health notification and any related case correspondence into a folder Cowork can read.
2. 2Add the relevant Performance Notifications and metric history (Order Defect Rate, Late Shipment Rate, etc.) as supplementary files.
3. 3Prompt: "Read everything in this folder. Identify the specific Amazon policy cited. Draft a Plan of Action with the four required sections. Pull supporting evidence from my files. Format the output as a clean PDF I can submit."

Cowork triggers the pdf skill, drafts the document, and gives you a submission-ready file in a few minutes. Compared to writing one from scratch — typically four hours when you're new to the process — this is the single highest-leverage Amazon use of Cowork. (If you'd rather skip the prompt engineering, the SellerForge POA Builder ships pre-trained on Seller Performance's current accepted format and updates each time Amazon shifts the rules.)

The honest caveat: the quality of the POA depends entirely on the quality of context you give it. Amazon's policy interpretations shift quarterly, the formats Seller Performance accepts vary by violation type, and the line between a warning and a full suspension is often unclear. A POA drafted by Cowork against a well-curated context folder is good. A POA drafted by Cowork against a vague prompt is generic and gets rejected. Treat this as a serious context-engineering task, not a chatbot question.

2. The Weekly Business Briefing (Scheduled Task)

Most operators spend their Monday morning hunting through five different Seller Central reports to figure out what happened the prior week. This is what scheduled tasks were built for.

The recipe:

1. 1Connect an Amazon Seller Central MCP server (more on this below) so Cowork can read your business reports.
2. 2Connect Gmail (or Outlook) so Cowork can email the output.
3. 3Schedule a task to fire every Monday at 7 a.m. that pulls last week's revenue, units, top movers, top decliners, account health status, and PPC spend; runs Claude's reasoning across the data to surface anomalies; and emails you a one-page briefing.

This is the kind of work that earns Cowork's keep. The first version takes an afternoon to wire up. After that, it runs on autopilot.

Two practical constraints. The scheduled task only fires when your computer is awake and the Claude desktop app is open — if your laptop sleeps overnight on a Sunday, Monday's briefing doesn't run. And each scheduled task burns rate-limit quota proportional to how much data it processes; a briefing that ingests 90 days of search-term reports each Monday will eat into your weekly cap quickly.

3. The Listing Audit Dispatch

Most Amazon listings are written once and never revisited. Cowork makes a good auditor:

1. 1Paste your ASIN into the prompt or upload your live listing as HTML/screenshots.
2. 2Prompt: "Audit this listing against Amazon's 2026 title and bullet best practices. Check character limits, keyword indexing rules, bullet hierarchy, and backend search term overlap. Flag specific weaknesses with revised copy options."

Cowork triggers the docx skill, returns a structured audit document with line-by-line suggestions, and you have a concrete edit list to take into Seller Central. It's not as deep as a purpose-built listing tool, but for sellers without one, this is meaningfully better than the existing listing. (See the SellerForge Listing Builder + Audit for the category-aware version.)

The limitation: Cowork doesn't know your category's actual top-ranking competitors or current keyword landscape unless you provide that data. It will give you a clean audit against general best practices. It will not give you the category-aware audit that compares your listing to the top three competitors actively indexing for your priority keywords. That's a different problem and a different tool.

4. The Reimbursement Triage Live Artifact

For sellers with even a modest catalog, Amazon owes more money than most realize — lost FBA units, damaged returns Amazon should have credited, fee overcharges, customer refunds without product returns. The eligibility windows are narrow and the evidence requirements are precise, which is why most sellers file maybe 10% of what they’re owed.

A Live Artifact in Cowork can help by acting as a recurring triage queue:

1. 1Connect an Amazon Seller Central MCP server with access to FBA inventory ledger, inbound shipment data, and reconciled financial events.
2. 2Build a Live Artifact that re-scans those reports each time you open it and surfaces a list of potentially reimbursable cases with case-type tags, date stamps, and dollar values.
3. 3From the artifact, fire a Dispatch on each case to draft the claim copy for submission.

The catch is that genuine reimbursement detection requires cross-referencing at least six different report types and applying Amazon's evolving reimbursement-policy windows (which materially changed in 2024 and again in early 2026). A Cowork artifact built against the basic reports will surface obvious cases but miss the harder ones — and the harder ones are usually the larger ones. We'll come back to this in the SellerForge section. (Or skip ahead: the SellerForge Reimbursement Claims module already does this cross-reference correctly.)

Connecting Cowork to Amazon: The MCP Server Question

Cowork can't talk to Amazon out of the box. There is no native Amazon connector in Anthropic's first-party MCP library. You have three options.

Option 1: Use a third-party Amazon MCP server. Several have emerged in 2026, with varying quality:

• MarceauSolutions amazon-seller-mcp (open source on GitHub) — covers core SP-API endpoints, includes a 2026 FBA fee calculator, OAuth authentication, and basic inventory optimization. Self-hosted.
• jay-trivedi/amazon_sp_mcp (open source on GitHub) — covers sales data, inventory, returns, and reports. Self-hosted.
• DataDoe Amazon Data MCP — managed service that exposes Seller Central, Vendor Central, Ads, inventory, fees, settlements, and profit data.
• agentcentral — hosted MCP server that connects Claude and other AI clients to Amazon Ads, Seller Central, inventory, orders, catalog, rankings, finance, and fulfillment data.
• Porter — sits between Amazon's SP-API and Claude, handling OAuth, rate limiting, and pagination so Claude only sees clean structured data.

The hosted options are easier; the self-hosted options give you more control. None of them handle every SP-API edge case, and the Advertising API in particular tends to be partial-coverage across the open-source servers.

Option 2: Build your own MCP server. Realistic if you're a developer or have one on staff. Wraps SP-API, handles auth refresh, normalizes the data, exposes a clean tool surface to Cowork. This is the cleanest long-term architecture but it's a meaningful engineering project — see the companion post on building an Amazon AI agent with Claude for the full architecture and cost breakdown.

Option 3: Skip the API and use browser automation. Cowork includes Claude in Chrome integration that can drive a real browser session against Seller Central. This works for tasks that don't need scale — pulling one report, drafting one response, downloading one file — and avoids the SP-API plumbing entirely. It will not work for production-grade automation because Seller Central's UI changes constantly, browser sessions expire, and the latency makes scheduled-task workflows impractical.

The pragmatic answer for most sellers: start with a hosted third-party MCP server for the read-only side (reports, business data, inventory) and use browser automation for one-off actions that need to be taken inside Seller Central. Move to a self-built server only when the volume justifies it.

The Skills and Plugins Worth Installing

The public plugin marketplace covers a lot of ground but very little is Amazon-specific. The high-value installs for an Amazon seller in 2026:

Anthropic's core skills (built in): docx, xlsx, pptx, pdf, and the schedule skill. Don't overlook these — most Amazon outputs are documents (POAs, audits, supplier briefs, deliverables) or spreadsheets (forecasting, reimbursement queues, keyword research). These four skills cover the output side of almost every workflow you'll build.

Gmail and Google Drive connectors: For email-driven workflows (sending the Monday briefing, processing supplier emails, archiving Amazon performance notifications) and for using Drive as a context library Cowork can reference.

Google Calendar: For workflows that depend on time (Prime Day prep timing, supplier lead-time calculations, promo windows).

An Amazon Seller Central MCP server of your choice from the options above. This is the single highest-value connector for Amazon work and the reason most sellers will care about Cowork at all.

A web search or research connector: For competitor monitoring and category research workflows. The native research tooling in Cowork is decent; a specialized connector (Tavily, Exa, or similar) handles structured retrieval better.

Custom skills you write yourself: This is where Cowork gets interesting. Open the skills folder and write a markdown file called amazon-poa.md that contains your specific instructions for how a Plan of Action should be structured for your category and your account's recent issues. Now every time you prompt Cowork about a POA, it pulls in those instructions automatically. Same pattern for your listing voice, your supplier brief format, your reimbursement claim templates, your weekly briefing format. Writing five or six of these custom skills is the difference between a generic Cowork setup and one that actually understands how you run your Amazon business.

Three or four hours invested in custom skills produces more lift than any plugin install. Most sellers never do it.

Where Cowork Stops Being Enough

After three months of running real Amazon workflows through Cowork, you'll notice the same set of friction points. They're not Cowork's fault — it's a general-purpose agent platform doing general-purpose things well. But they matter if you're trying to run an Amazon business on it.

Context is yours to manage. Cowork doesn't natively understand Amazon's policy taxonomy, the difference between an ODR violation and a Used Sold as New claim, the specific language Seller Performance currently wants to see in a Section 3 appeal, or the way Amazon's 2026 reimbursement policy window narrowed from 90 days to 60. Every one of those nuances has to be loaded into your prompts or your custom skills. When Amazon changes the rules — which happens quarterly — your skills are out of date until you update them.

No native Amazon data model. Cowork sees JSON from an MCP server; it does not know what the JSON means. Your top-3-organic-rank keyword overlapping your highest-spend PPC keyword is a $400/month cannibalization problem — but only if the agent on top of the data knows to ask the question. Cowork won't ask it unless you prompt it to. Purpose-built tools ask it by default. (See Beyond ACoS: The Advertising Metrics That Actually Matter in 2026 for the cannibalization framing in detail.)

Scheduled tasks need your laptop awake. This is the limit that bites soonest. The 7 a.m. Monday briefing doesn't run if your computer was asleep at 6:59 a.m. You can work around this (a small always-on Mac mini in your office, the Claude mobile app's dispatch surface, a self-hosted version of the agent), but each workaround adds friction.

Rate limits compound with sub-agents. Cowork tasks routinely spawn sub-agents and tool calls. A single complex session can consume as much weekly quota as dozens of chat messages. Anthropic permanently doubled the 5-hour rate limits in May 2026 and added pay-as-you-go overflow for Pro and Max plans, but if your workflow involves nightly briefings, weekly reimbursement scans, and ad-hoc Dispatches, you will hit the weekly cap on a Max 5x ($100/month) plan and have to either upgrade to Max 20x ($200/month) or take overflow billing at API rates.

No persistent Amazon-specific intelligence. Cowork doesn't track your category's competitor pricing history, your ASINs' Buy Box win rate over time, your search-term harvest opportunities, or your account-level performance trends. Each session starts fresh against whatever data you load into it. You can fake persistence with Projects and a brain folder, but you're building the intelligence layer yourself, one skill at a time.

Amazon's APIs change without notice. When an SP-API endpoint deprecates, your MCP server breaks. When Amazon's Advertising API restructures a report format, your scheduled task starts surfacing wrong numbers. Maintenance is your job. For most sellers running Cowork, this maintenance is the silent cost that erodes the time savings over a year.

None of this means Cowork is the wrong tool. It means Cowork is the right tool for general work and a partial tool for Amazon work. Knowing the difference saves you months.

SellerForge: Cowork's Amazon Layer, Done

Everything described above — the SP-API integration with proper auth refresh, the Ads API connection with report polling and rate-limiting, the listing audit that actually compares against top-ranking competitors, the Plan of Action generator trained on Amazon's current policy language, the reimbursement detection logic that cross-references six report types, the forecasting that pulls sales velocity and ad spike data and supplier lead time into one calculation, the persistent account intelligence that tracks your performance over time, the always-on monitoring that doesn't depend on your laptop being awake — that's what SellerForge is.

SellerForge runs the same underlying Claude models you’d use in Cowork (Sonnet for routine analysis, Opus for the hardest reasoning, Haiku for cheap classification under the hood), but the entire stack above the LLM is purpose-built for Amazon and continuously updated for Amazon’s policy and algorithm changes.

The specific differences that matter for an Amazon seller:

Speed. A SellerForge listing audit returns in 12 seconds. The Cowork equivalent — assuming you've already wired up an MCP server, written custom skills, and loaded category context manually — takes minutes per ASIN. Across a 40-SKU catalog the math is brutal.

Specialized training on Amazon's current state. Amazon's policies, fee structures, algorithm signals, and report formats change quarterly. SellerForge's prompt library, scoring models, and detection logic are updated for each change as it ships. Your Cowork skills are updated when you remember to update them.

Business-data context loaded by default. SellerForge ingests your Seller Central account on connect — last 12 months of sales, inventory, advertising, account health, reimbursements, returns — and that context is present in every conversation. In Cowork, you build that context layer one prompt at a time.

Edge cases handled. The reimbursement module catches the cases the basic open-source MCP servers miss because it cross-references the six report types correctly and applies the current policy windows. The Ads module diagnoses cannibalization (paying for clicks on keywords where you already rank organically) by joining ad data and Brand Analytics data — a join Cowork doesn't make unless you write the skill yourself. The POA Builder writes against Seller Performance's current accepted format, not last year's.

Always on. SellerForge runs on hosted infrastructure with continuous monitoring of your account. Performance Notifications, inventory thresholds, account health changes, and price drops trigger alerts whether your laptop is awake or not. Scheduled tasks fire on time, every time.

Audit trail and traceability. Every recommendation SellerForge makes is traceable to the underlying data. When SellerForge says "Your ACoS jumped because a new competitor launched at $19.99 four days ago," you can click through to the Keepa price history, the Buy Box win-rate trend, and the timestamp. Building that traceability layer on top of Cowork is its own multi-week project.

A useful mental model: Cowork is the engine; SellerForge is the car. You can drive somewhere with just the engine if you really want to, and developers do it every day for the fun and the learning. For everyone else, the wrapper is the product.

Claude Cowork vs. SellerForge: An Honest Comparison

The honest TL;DR: at $99/month, SellerForge costs less than a single Max 20x Cowork plan, and it does the Amazon-specific work Cowork won't do unless you build it yourself. See the full module list for everything that's included.

When Cowork Is Still the Right Answer

Despite all of the above, there are legitimate reasons to run your Amazon workflows in Cowork instead of (or in addition to) SellerForge:

You're a developer who genuinely enjoys building the stack. Cowork is a wonderful sandbox for learning MCP, agent design, and LLM application patterns. Building your own Amazon agent is one of the better self-directed projects for picking up modern AI tooling. (Again — see the companion architecture post.)

You have requirements outside Amazon. SellerForge is purpose-built for Amazon. If your operations span Shopify, Walmart, eBay, and Amazon, and you want one assistant across all of them, a general-purpose Cowork setup with multiple marketplace MCP connectors is the more flexible architecture. (We'd argue you still want SellerForge for the Amazon-specific work and Cowork for everything else — they coexist well.)

Your workflows are highly idiosyncratic. Custom internal reporting formats, proprietary supplier integrations, unique compliance workflows — these are the cases where the customizability of Cowork outweighs the speed of a purpose-built tool.

You're a single seller experimenting and not yet ready to subscribe to anything. The Cowork Pro plan at $20/month is cheap enough to learn on. If you're not yet sure your Amazon business needs more than ad-hoc AI help, start there.

If you're going to run Amazon work through Cowork, here's what I'd tell you:

• Write custom skills before you write any prompt twice. The second time you prompt Cowork to draft a Plan of Action, that should be the moment you create amazon-poa.md in your skills folder.
• Use Projects, not loose chats. One project per brand or per account. Loose chats lose context fast.
• Build your brain folder before your first complex workflow. A CLAUDE.md file with your account context, your brand voice, your fulfillment model, your top SKUs, and your current account-health status is the single biggest accuracy improvement you can make.
• Set up data-freshness checks on every scheduled task. Silent failures are the worst kind. If the Monday briefing has stale data because the MCP server broke on Wednesday, you want to know about it before the briefing emails you a wrong answer.
• Watch your rate limits. Cowork burns quota much faster than chat. Monitor weekly usage and budget your scheduled tasks accordingly.
• Treat policy changes as a maintenance category. When Amazon updates a policy, your custom skills are out of date until you update them. Schedule a quarterly review of every skill against current Amazon documentation.

The first 80% of Amazon work in Cowork is genuinely fun. The last 20% is months of edge-case maintenance. Know that going in.

Closing

Claude Cowork is one of the better general-purpose AI agent platforms shipping in 2026. For Amazon sellers willing to invest the setup time, it is a meaningful productivity multiplier. For Amazon sellers who want the same outcomes without the setup time, the MCP plumbing, the custom-skill maintenance, and the rate-limit management — that's what SellerForge was built for.

If you want to see the difference in practice, start a free SellerForge trial and connect your Seller Central account. The Listing Builder, the POA Builder, the reimbursement queue, the forecasting module, the Ads diagnostics, and the always-on monitoring are already running. The setup is five minutes, not a weekend.

Share this article

Facebook

LinkedIn

X

ThreadsCopy link