Example output: top product search results for yoga mat on Amazon US, with price, reviews, and BSR.You just cloned the repository. The install script finished. Your AI agent has discovered the tools. Now what.
This guide walks through the first five minutes: what to verify, what to ask, and how to know it is working. No setup theory. No architecture diagrams. Just the commands that prove the pipeline is live and the questions that produce useful answers.
After running python3 scripts/install.py, restart your AI agent. Then confirm the tools loaded correctly. The exact command depends on your agent:
Claude Code:
/mcp
Example output: top product search results for yoga mat on Amazon US, with price, reviews, and BSR.
This lists every MCP server and its tools. Look for sorftime-seller-agent with a tool count to its right. If it appears, the installation succeeded.
Codex CLI:
/mcp list
Cursor:
Open the MCP panel from the settings menu and check that sorftime-seller-agent shows as connected.
If the tools do not appear, the most common cause is that the MCP key was not saved. Run python3 scripts/install.py again and verify the key was entered correctly at the prompt. The key is available free at open-intl.sorftime.com.
Do not start with a complex multi-step analysis. Pick a single, concrete question that hits one tool. This confirms the data pipeline works end to end and gives you a baseline for the response format.
Search "yoga mat" on Amazon US and show me the top 3 results with price, review count, and BSR.
Your AI agent calls product_search, receives structured data, and returns a summary. You should see actual ASINs, real prices, and current review counts — not placeholder data. That distinction matters. If the agent returned numbers from its training data, the BSR values would be stale. If it called the MCP tool, they reflect the live marketplace.
A successful response looks something like this (the actual data will differ):
Top 3 results for "yoga mat" on Amazon US:
1. B0XXXXXXX1 — $24.99 — 12,400 reviews — BSR #48 in Sports & Outdoors
2. B0XXXXXXX2 — $32.50 — 8,200 reviews — BSR #112 in Sports & Outdoors
3. B0XXXXXXX3 — $18.99 — 5,900 reviews — BSR #203 in Sports & Outdoors
If you see this, the pipeline is working. Move on.
Now stack another layer onto the same query. Cross-reference the top results with keyword intelligence.
For the top "yoga mat" result on Amazon US, what keywords is it ranking for? Show me search volume and estimated organic position for each.
Your agent now chains two tools: the product detail tool to confirm the ASIN, and the keyword tool to pull its traffic terms. The response reveals which search terms drive the listing’s visibility and how it ranks for each.
This is where the value of MCP becomes visible. In a manual workflow, this step requires switching from a product search tool to a keyword research tool, copying an ASIN from one to the other, and waiting for separate queries to resolve. Here, the agent handles the chain. You asked one follow-up question.
Individual ASIN analysis is useful. Category-level analysis is where things get interesting. Move from a single product to the broader market:
Give me a category overview for "coffee makers" on Amazon US. Include the average price, brand concentration, review distribution among the top 20 listings, and estimated monthly search volume for the main category keywords.
Your agent chains category_report for market structure, product_search for pricing and review data across the top listings, and keyword_detail for demand signals – then synthesizes the results into a single answer.
What you get back:
The agent correlates signals across tools that were never designed to talk to each other. That cross-dimensional synthesis is what makes conversational access different from clicking through individual reports.
Data without cost context is incomplete. Pick one product from the previous response and run a profitability estimate:
For the ASIN [paste one from the response], calculate estimated FBA fees and breakeven at a $24.99 selling price with $7.50 estimated landed cost, assuming standard-size shipping and 3% return rate.
Your agent calls the FBA fee calculator and profit estimator tools, then returns a structured breakdown:
This is a modeled estimate, not a guarantee. Marketplace fees change, shipping costs vary, and actual returns may exceed projected rates. The value is having a structured baseline in under a minute instead of building a spreadsheet from scratch.
The last step in the quickstart is moving from one-off queries to ongoing monitoring. Ask your AI agent to remember a recurring check:
Every Monday morning, check if any new listings have entered the top 20 for "stainless steel water bottle" on Amazon US. If a new competitor appears, show me their ASIN, price, estimated monthly sales, and review count.
This leverages two things: the MCP tools for product discovery at scale, and your AI agent’s ability to schedule recurring tasks. The tools pull fresh data each time. The agent formats the report. You read it over coffee.
In five minutes, the following took place:
| Action | Tools Called | Manual Equivalent |
|---|---|---|
| Verified installation | MCP registry query | N/A |
| Simple product search | product_search |
Browse Amazon, note ASINs |
| Keyword cross-reference | product_detail + keyword_detail |
Switch between two separate tools |
| Category market overview | category_report + product_search + keyword_detail |
Open three dashboards, cross-reference manually |
| Profitability estimate | FBA fee + profit calculator | Build a spreadsheet from scratch |
| Recurring competitor watch | Scheduled product_search with filters |
Remember to check manually every week |
None of these steps required opening a browser tab, copying an ASIN between tools, or remembering to check something next Monday.
The sorftime-seller-agent handles the tool-calling layer. You handle the questions.
Tool discovery fails after install. Run the install script again and check the MCP key was saved. Restart your AI agent. If you are on Cursor, verify the MCP server path in settings points to the correct Python environment.
The agent answers from training data instead of calling tools. This is a common issue on the first query. Ask explicitly: “Use the MCP tools to search…” — this signals to the agent that it should call external functions rather than rely on stored knowledge.
Slow responses on the first query. The first call to any MCP tool involves loading the Python environment and establishing an API connection. Subsequent queries are faster. If every query is slow, check your network connectivity to the Sorftime API endpoint.
Rate limits or key errors. Free-tier keys have usage limits. If you hit them, the agent will surface an error message indicating the limit. Paid plans with higher limits are available at open-intl.sorftime.com.
The five-minute path above covers product search, keyword analysis, category intelligence, profitability estimation, and recurring monitoring. Each of those is a surface-level use of deeper tools. As you work with the agent, you will discover additional capabilities:
The sorftime-seller-agent is open source. You can inspect every tool definition, understand the data sources, and adapt the workflow to your business. The repository includes example scripts and documentation for extending the tool set.
Start here.
git clone https://github.com/DannylydST/sorftime-seller-agent
cd sorftime-seller-agent
python3 scripts/install.py
A free MCP key is available at open-intl.sorftime.com. No credit card required.
After installation, open your AI agent and ask the first question from Minute 1 above:
Search "yoga mat" on Amazon US and show me the top 3 results with price, review count, and BSR.
The tools are wired up. The rest is conversation.
References
[1] Sorftime Seller Agent GitHub Repository — https://github.com/DannylydST/sorftime-seller-agent [2] Sorftime International MCP Portal — https://open-intl.sorftime.com [3] Model Context Protocol Specification — https://modelcontextprotocol.io —