Methodology

What we track

The record currently covers 54 tools for Amazon and marketplace sellers across research, PPC, profit analytics, repricing, listing content, reviews, inventory, multi-channel and data APIs. For each tool we keep three addresses: the pricing page, the changelog or release notes, and the help center. We check that each address answers before it enters the list, and we record what we found.

Inclusion is not an endorsement. A tool is on the list because sellers use it or ask about it, not because it pays us. How we earn money explains the commercial side.

Daily checks

• We fetch each pricing page and changelog with a crawler that identifies itself as SellerToolsIndexBot and respects robots.txt. Requests to the same domain are spaced several seconds apart.
• We store the visible text of the page, not the design. A new snapshot is saved only when the text changed.
• The first snapshot of a page is a baseline. Changes are logged from the second check onward, as one of five event types: price string changed, plan added, plan removed, page text changed, or a line that mentioned a feature disappeared.
• A page whose text suddenly shrinks below 40 percent of the last snapshot is not saved. That pattern usually means a bot wall or an error page, not a real change.
• Dates, times and copyright years are masked before comparison, so a footer that says 2026 instead of 2025 does not count as a change.

An event from this process means the page changed. It does not yet mean a price went up. Until an editor has compared the two versions, the change log shows it with an unverified label.

Archived history

To see further back than our own checks, we read archived copies of each pricing page from the Internet Archive's Wayback Machine, roughly one per quarter for the last three years. Every data point keeps the link to the exact archived copy, so you can open the page as it looked on that day. So far that is 297 archived pricing pages for 40 tools, the oldest from Oct 31, 2023.

Our requests to the archive are throttled and identify themselves honestly. Some vendors render prices with scripts that the archive did not capture; for those quarters you will see Not published rather than a number.

Machine-read and verified prices

Pricing pages are written for people, not machines. Plans are named differently, annual and monthly prices sit next to each other, add-ons look like plans. Software can read the strings, but it gets some of them wrong. So we separate two kinds of numbers:

The entry price is the cheapest plan billed monthly, in US dollars, that a single seller can buy without a sales call. Annual prices are shown separately as a monthly equivalent. When a vendor does not publish a price, the cell says Not published. When we have not checked it yet, it says Not verified. We do not estimate.

STI Cost Index

The STI Cost Index answers one question: did seller software get more expensive for you this quarter? It follows the method statistics offices use for the smallest building blocks of a consumer price index.

1. Input. Only verified entry prices. Each one is checked individually against the archived page (or the live page for the current quarter) and stored with the quoted text that shows the price and the plan name. Checks are done by our research agent and spot-checked by the editor. Machine-read prices never enter the index.
2. Quarterly price per tool. The latest verified price inside the quarter. A price stays valid until the next verified observation of the same tool, and only between the first and the last observation we have for that tool.
3. Withdrawn plans. We track the cheapest self-serve plan billed monthly. When a vendor drops that plan, the index follows the new cheapest plan, because that is what entry now costs you. Renames at the same price do not move the index. Every plan switch is named in the tool's price record.
4. Quarter to quarter. For every tool priced in both quarters we take the ratio of the two prices. The index moves by the geometric mean of those ratios (a Jevons index). Tools enter and leave the sample without distorting the level, because each step only compares a tool with itself.
5. Chaining. The steps are multiplied together. The first published quarter equals 100.
6. Minimum sample. Every quarter-to-quarter step must rest on at least 8 tools, and the series must cover at least 4 quarters.

Status on Oct 31, 2026: published, base 2024-Q1. Verified prices on file: 137. Archived pricing pages waiting for review: 297.

The index is not weighted by market share, because no vendor publishes subscriber counts we could verify. Each tool counts once. A price cut in one tool and a rise of the same proportion in another cancel out.

MCP servers and agent access

MCP servers let an AI agent act in your seller or ads account. The question that matters is what the agent can change. We have 97 servers on record with 2,596 tools between them, 650 of which we mark as able to write.

• Open-source servers are found through GitHub search. We clone the repository and extract every tool definition from the code. Where we can run a server without your credentials, we call tools/list and record the live answer.
• Hosted vendor servers usually hide the full tool list behind a login. For those we read the public documentation and mark the tool list as partial.
• A tool counts as write when it creates, updates, deletes, cancels or submits something. Everything else is read.

Every access mark carries its basis. Today most marks rest on the tool's name and description, shown as name based. A tool called get_orders almost certainly reads; a tool called sync_inventory might do either. We will replace name-based marks with tested ones as we install servers in a sandbox account.

SP-API watch

Most seller tools depend on Amazon's Selling Partner API. When Amazon removes an endpoint or a report type, tools break or lose features. We follow four primary sources: the amzn/selling-partner-api-models repository on GitHub, the SP-API deprecation schedule, the SP-API release notes and Amazon's developer fee pages. 299 events are on record. Each event keeps two dates: when the change takes effect, which can lie in the future for scheduled removals, and when we detected it.

Chrome extension signals

For 44 browser extensions we read the public Chrome Web Store listing: user count, rating, number of ratings, version and last update. Each extension is matched to its tool by a link from the vendor's website, the developer website in the store, or the developer name, and we record which one. User counts in the store are rounded by Google and are a rough signal, not a market share.

Colours and labels

Colour on this site always means something:

• red worse for you: a price went up, a limit went down, an agent can change data.
• green better for you: a price went down, an agent can only read.
• amber something was taken away: a feature, a plan, an API.
• Yellow highlight marks the evidence in a sentence.

Everything else is black on paper. We do not use vendor logos, and we do not use Amazon's.

Corrections

If you find an error, write to info@sellertoolsindex.com with the page and, if you can, a link to the source. Vendors are welcome to do the same. We correct facts quickly and note the correction on the page. We do not remove accurate information on request.