Connecting Amazon Seller Data to Claude: Step-by-Step Guide

Choose the Right Connection Method

None of these methods is automatically low risk. Privacy depends on the data, account settings, permissions, storage and logs across every service in the workflow.

Method 1: Upload Amazon Reports to Claude

Step 1: Pick One Question and the Right Report

For a first review, ask: “Which ASINs have meaningful traffic but weak sales performance?” Export the relevant Detail Page Sales and Traffic report from Seller Central Business Reports. Record the marketplace, date range and ASIN level.

For search visibility, use Search Query Performance where your account is eligible. For advertising spend and attributed sales, export the appropriate Amazon Ads report. Keep these sources separate until their identifiers, dates and metric definitions are aligned.

Step 2: Clean the File Without Removing Useful Evidence

1. Keep an untouched original. Work on a copy with one header row and consistent column names.
2. Move explanatory text and totals out of the data table. Retain totals separately for reconciliation so they are not counted as another ASIN.
3. Keep zero-sales rows. They may reveal wasted spend, weak conversion or an inactive product. Flag missing values rather than replacing them with zero.
4. Label currency, timezone, reporting dates and whether rows represent parents, children or SKUs. Preserve identifiers as text.
5. Remove unnecessary buyer names, addresses, phone numbers and other personal information. Confirm that you are allowed to process the remaining data.
6. Save a CSV. XLSX uploads require Claude's code execution and file creation capability to be enabled; use CSV if that feature is unavailable.

See Claude's file-upload guidance for supported formats and current limits.

Step 3: Create a Project or Start a Chat

Create a project for the brand if you want reusable instructions and reference files. Add verified product specifications, brand guidelines and metric definitions. Attach the dated performance report to the analysis conversation, or add it to project files if supported. Remove or clearly label superseded reports so periods are not mixed.

For a one-off review, a regular chat with the report attached is sufficient. Neither method grants access to your Amazon account.

Step 4: Give Claude a Specific Analysis Prompt

Review the attached Amazon reports for the stated marketplace and dates.

First list the files, periods, row counts and available metrics.
Flag missing columns, duplicate identifiers and incompatible reporting periods.
Use the exact metric names from each source.
For Business Reports, calculate Unit Session Percentage as
units ordered / sessions x 100 when both inputs are available.
Do not treat that metric as the same as advertising conversion rate
or Search Query Performance purchase rate.
Keep zero-sales rows. Flag zero denominators instead of dividing by zero.
Use code for calculations if available and show the formulas.
Separate observed facts, possible explanations and recommended checks.
Return the ASIN, supporting numbers, issue and next action.
Do not invent product features or claim that a correlation proves a cause.

Step 5: Verify Before Using the Recommendations

Compare totals and several ASIN-level calculations with the original export. Ask Claude to identify the rows behind each finding. Investigate stock availability, price changes, traffic mix and listing issues before assuming weak performance is caused by copy.

For listing rewrites, provide the current listing and verified product facts. Sales data can identify a product worth reviewing, but it cannot establish its materials, certifications, compatibility or benefits.

Method 2: Build a Scheduled Low-Code Workflow

Once the manual review works, a tool such as Make or n8n can orchestrate the same process. This is an implementation outline: connector coverage and configuration vary, so verify the exact operations your chosen tool supports.

1. Choose the data source. Use a dated file placed in an approved storage location, or an authorized Amazon integration. Do not assume every Seller Central report can be emailed automatically or retrieved by every connector.
2. Retrieve and validate the data. Check the account, marketplace, period, report status and row count. Stop and alert the operator if data is missing or stale.
3. Prepare a compact input. Calculate totals and rates in code or spreadsheet formulas. Send relevant metrics, exceptions and definitions to Claude rather than an unexplained raw export.
4. Configure Claude access. Create an API key in the Claude Console and store it in the automation platform's credential store. Confirm API billing and model access separately from your Claude chat subscription.
5. Call the model. Use the platform's supported Claude module or implement the Messages API. For direct HTTP, follow the current authentication and version-header requirements. Choose an available model from the current documentation and keep its ID configurable.
6. Validate the response. If the next step expects JSON, validate it against a schema. A prompt asking for JSON alone is not a guarantee. Reject incomplete responses and handle errors before continuing.
7. Deliver a reviewable summary. Include source dates and missing-data warnings. Route it to your chosen tracker or internal channel, with retries and duplicate protection.

Check the model lifecycle page when building and maintaining the workflow. Retired models will not serve requests.

For inventory alerts, define days of supply explicitly. A simple estimate is sellable units divided by average daily units sold. Flag stockout periods, seasonality and zero sales; assess inbound stock and arrival dates separately. A fixed 21-day threshold is an operational choice, not a universal replenishment rule.

Method 3: Build a Custom Amazon API Integration

This route requires a developer to implement authentication, report processing, retries and access controls. Claude analyzes the data your application supplies; it does not acquire Amazon permissions by itself.

Step 1: Register and Authorize the Application

Follow Amazon's current SP-API onboarding instructions for your application type. Request the roles needed for the data you intend to retrieve, register the application and obtain the appropriate seller authorization.

Private applications use self-authorization; public applications use the applicable OAuth authorization flow. Your application uses Login with Amazon credentials and the authorized refresh token to obtain access tokens. Protect these secrets and never include them in prompts.

Amazon removed the AWS IAM and Signature Version 4 requirement for SP-API calls in October 2023. Separate AWS services used by your infrastructure can still require their own permissions.

Step 2: Request the Correct Seller Report

For seller Business Reports, use GET_SALES_AND_TRAFFIC_REPORT. It returns JSON and requires the relevant Brand Analytics API role. Set marketplace IDs, reporting dates and the date/ASIN granularity explicitly. A vendor report is not a substitute for seller data.

For SQP, use GET_BRAND_ANALYTICS_SEARCH_QUERY_PERFORMANCE_REPORT where eligible. Amazon requires the Brand Analytics role and Brand Registry enrollment. Use the required ASIN and report-period options with valid weekly, monthly or quarterly boundaries. Scheduling your application daily does not turn SQP into a daily report.

Review Amazon's analytics report reference for eligibility, options and availability.

Step 3: Wait for Processing and Download the Document

1. Call createReport with the selected report type and parameters. Save the returned reportId.
2. Check getReport(reportId) with bounded polling and backoff, or use report-processing notifications. Handle CANCELLED and FATAL explicitly.
3. When processing is DONE, read the reportDocumentId from the report metadata.
4. Call getReportDocument(reportDocumentId). The report ID and document ID are different identifiers.
5. Download from the returned pre-signed URL before it expires. Decompress if the response specifies compression, then parse the report's actual format.
6. Validate the account, reporting period and totals before sending selected data to Claude.

Amazon's report retrieval instructions describe the download response. Do not treat document metadata as the report contents.

Step 4: Analyze First, Add Write Operations Separately

Start with read-only reporting. Send validated metrics to Claude through the Messages API and store its recommendations alongside the source data and reporting period.

For listing changes, use the appropriate SP-API Listings Items operations and validate attributes against the applicable product type requirements. For advertising reports, bids and budgets, use the separate Amazon Ads API with its own access and authorization process.

Before enabling writes, require approval of the exact proposed change, retain the previous value and log the API response. Recheck the current value before submitting so an old recommendation does not overwrite a newer decision.

Privacy: Training and Retention Are Different Questions

For consumer Claude accounts, model improvement depends on your choices and the policy's exceptions, including feedback and safety review. Check the consumer training policy and your privacy settings before uploading business data.

Anthropic says it does not use commercial-product inputs and outputs for training by default. That does not mean data is never retained or reviewed. Standard API inputs and outputs are generally deleted within 30 days, subject to exceptions; services such as saved chats or the Files API have different retention behavior. Zero data retention requires applicable terms and has product/model limitations.

Read the current commercial training policy and retention policy. Confirm what applies to your specific account, model and features.

• Minimize data before upload. Performance analysis usually does not need buyer identity fields.
• Confirm client authorization and applicable Amazon data-handling requirements. Removing names alone does not establish compliance.
• Review middleware logs, file storage, sharing and retention as well as Claude settings.
• Keep API keys and refresh tokens in credential stores or secret managers, outside reports, prompts and repositories.
• Treat report contents as data, not instructions. Validate model output before another system acts on it.