You have run every keyword tool. The dropdown menus, the reverse ASIN lookups, the search volume estimators. You are bidding on the same keywords as everyone else in your category. And your ACOS keeps climbing.
This is not a failure of keyword research. It is a structural limitation of the tools. Every seller in your category opens the same dashboard, types the same seed keyword, and gets roughly the same list back. The output converges. Then the bids converge. Then the margins disappear.
The keywords that actually move the needle – the ones with real search volume and low competition – are not in that list. They are one or two degrees removed from the obvious terms. Finding them requires looking at the data from a different angle.
Most Amazon keyword tools work the same way. They take a seed keyword or a competitor ASIN, look up which search terms drive traffic, estimate monthly volume, and rank them by relevance. The algorithm is deterministic. Given the same inputs, every seller gets the same output.
The problem is not that the data is wrong. It is that everyone is acting on the same signal. When three hundred sellers all see the same “high volume, medium competition” keyword and add it to their campaigns, the competition goes up. The bid goes up. The ACOS goes up. The keyword was genuinely good when the tool surfaced it. It stopped being good the moment everyone saw it.
The keywords that stay profitable are the ones the tools deprioritize – the long-tail variants, the intent-rich phrases with lower reported volume that actually convert at twice the rate, the seasonal terms that spike for eight weeks and then disappear from the default time window. Finding them is not about having a better keyword list. It is about asking better questions of the data.
Model Context Protocol (MCP) changes how data gets queried. Instead of opening a tool and clicking through pre-built reports, you connect your AI agent directly to marketplace data and ask questions in plain English. The agent queries the data, applies your criteria, and returns results that match your specific logic – not a generalized ranking algorithm that every other seller in your category is also looking at.
The sorftime-seller-agent is an open-source MCP server that exposes marketplace intelligence as tools an AI agent can call. It covers keyword research, competitive analysis, category scouting, and performance monitoring across multiple marketplaces. The difference is not in the data – it is in how the data gets filtered.
A dashboard shows you what its designer thought was important. An MCP-connected AI agent shows you what you asked it to find.
Here is a concrete example. Suppose you sell kitchen storage products on Amazon US. The obvious keywords – “kitchen storage,” “food storage containers,” “pantry organization” – are saturated. Every competitor bids on them. The following approach uses the sorftime-seller-agent MCP to surface keywords that other sellers are likely missing.
First, install the MCP server and connect it to your AI agent:
git clone https://github.com/DannylydST/sorftime-seller-agent
cd sorftime-seller-agent
python3 scripts/install.py
Once connected, the prompt below asks the AI agent to run a multi-step keyword discovery workflow that goes beyond any single tool’s default output:
*Example output: hidden long-tail keywords with search volume, competition, and bid estimates.*
Analyze the top 20 ASINs in "kitchen storage" on Amazon US. For each ASIN,
retrieve the organic and sponsored keywords it ranks for in positions 1-20.
Then:
1. Filter to keywords where fewer than 3 of the top 20 ASINs rank -- these
are likely missed by competitors doing standard reverse-ASIN research.
2. Cross-reference those keywords against monthly search volume from the
keyword research tool. Keep only keywords with >500 monthly searches.
3. Check the review velocity of the top 3 organic results for each keyword.
Flag keywords where the top-ranking products have under 200 reviews --
indicating lower competitive defensibility.
4. For the keywords that pass all three filters, estimate the suggested bid
and compare it to the category average CPC. Highlight keywords where the
suggested bid is below the category average by at least 20 percent.
Output a table with: keyword, monthly search volume, number of top-20 ASINs
ranking for it, average reviews of top 3 organic results, suggested bid,
category average CPC, and a one-line verdict on whether to target it.
The agent runs the queries sequentially, applying each filter to the output of the previous step. The result is a shortlist of keywords that meet specific strategic criteria – not a ranked list of whatever the tool’s algorithm decided to show.
The workflow above catches keywords that standard tools systematically underreport. They fall into three categories:
Intent-qualified long-tail keywords. Search terms like “airtight cereal container for small pantry” have lower reported search volume than “cereal container,” but the buyer intent is far higher. Someone searching for that term knows exactly what they want. Standard tools rank these terms low because their volume looks unimpressive in isolation. But when filtered by conversion signal rather than raw volume, they become high-value targets.
Adjacent-category keywords. A kitchen storage product might also rank for “camping food container” or “meal prep containers for gym.” These terms sit in adjacent use cases that the product fits but that most sellers in the category never research because they are not obvious from the category page. The cross-referencing step in the prompt catches them because it scans what all top ASINs rank for, including terms outside the obvious category boundaries.
Seasonal and trend-spike keywords. A keyword like “pantry organization for holiday baking” spikes in October and November, then drops. If the tool’s default time window is 30 or 90 days, you will not see the spike unless you happen to check during those months. Querying the data with a custom time window – or asking the AI agent to compare month-over-month trends – surfaces these terms before they become obvious to everyone else.
The common thread: these are not obscure keywords. They are keywords that the standard tool workflow de-emphasizes because they do not fit the one-size-fits-all ranking model.
The contrast is straightforward. The manual keyword research workflow goes: open tool, enter seed keyword, scroll through results, add promising terms to a spreadsheet, open another tool to verify volume, manually estimate competition by glancing at review counts on the SERP. Performed thoroughly, this takes 30 to 45 minutes per product. Most sellers do a faster version and miss the edges of the keyword landscape.
With MCP: type the prompt. Wait 30 to 60 seconds. Read the shortlist. The time savings are real, but the competitive advantage is in the filtering logic. You are not just searching faster. You are searching differently – asking questions the pre-built reports cannot answer.
It does not guarantee low ACOS. Amazon advertising performance depends on listing quality, pricing, reviews, and a dozen other factors beyond keyword selection. Finding good keywords improves the odds. It does not eliminate the work of testing and optimizing.
It does not replace human judgment. The AI agent can filter keywords by your criteria, but setting the criteria – which metrics matter, what thresholds to use, which trade-offs are acceptable – is still a human decision. The tool handles the reading. The seller handles the thinking.
It requires an MCP-compatible AI agent. The sorftime-seller-agent works with Claude Code, Codex, Cursor, and other MCP clients. It is not a standalone web application. If you are not already using an AI coding agent, there is a one-time setup step.
The sorftime-seller-agent is open source. Install it, connect it to your AI agent, and start asking questions that go beyond the default reports.
git clone https://github.com/DannylydST/sorftime-seller-agent
cd sorftime-seller-agent
python3 scripts/install.py
To use the MCP tools, create a free account at open-intl.sorftime.com for API access. The free tier covers keyword research, ASIN analysis, and category scouting. No credit card required.