Sif Amazon Research

Sif Amazon Research

Operating rules

Use Sif MCP as the evidence source whenever available. Do not infer final business conclusions from one metric alone.

Before calling tools, identify the task type:

• Market/opportunity research: keyword or category direction.
• Market validation: keyword + candidate ASIN viability.
• Competitor analysis: competitor ASIN growth path and traffic source.
• Keyword layout: main terms, long tails, root structure, missing terms.
• Launch evaluation: whether to push a product.
• Post-launch feedback: continue, optimize, or stop.
• Growth optimization: expand proven traffic and keywords.
• Traffic/ad diagnosis: root cause of traffic or ad performance changes.

If the request does not include enough identifiers, ask for the missing ASIN, keywords, country, and time window. Default country to US only when the user does not specify a marketplace.

Core workflow

1. Define the decision being made: enter / not enter, launch / not launch, scale / optimize / stop, or diagnose root cause. 2. Collect Sif evidence across the minimum necessary domains:

• Market demand: keyword demand, history, root trend.
• Competition: keyword competition, top ASIN concentration, competitor keyword signals.
• Traffic: ASIN/listing traffic trend, structure, keyword distribution.
• Sales: sales list/trend, variant performance.
• Ads: ASIN ad structure, ad trend, campaign contribution, campaign/ad group drill-down.

3. Cross-check evidence. Separate confirmed findings from hypotheses. 4. Produce a business decision with confidence, evidence chain, risks, missing data, and next validation actions.

For scenario-specific tool sequences, read references/research-playbooks.md. For tool selection and parameter conventions, read references/sif-tool-map.md. For real-world MCP response quirks, large-output handling, and field interpretation pitfalls, read references/field-notes.md before any full analysis or debugging.

Required output format

For every Sif-based research answer, include:

1. Decision / conclusion: the direct answer and confidence level. 2. Problem layer: market demand, competition, traffic, ads, product/listing, or mixed. 3. Evidence chain: concise bullets linking each claim to Sif data returned by tools. 4. Business impact: what the evidence means for sales, traffic, ranking, or launch risk. 5. Priority actions: the top 3 actions, ordered by expected impact and speed. 6. Do not do yet: actions blocked by missing or weak evidence. 7. Missing data: 3 data gaps and which conclusion each gap affects. 8. 7-day validation plan: measurable checks the user can run quickly.

If data is incomplete, say what current data supports, what is missing, and whether the recommendation is only a hypothesis or execution-ready.

Guardrails

• Do not conclude from ACOS alone, CTR alone, or sales change alone.
• Do not output a final plan when key Sif data is unavailable; label it as a hypothesis.
• Do not treat analyze_traffic_anomaly as a generic raw-data tool. Use it only for explicit traffic anomaly/root-cause diagnosis or when the user explicitly asks to diagnose traffic.
• Preserve any render_footer returned by Sif tools verbatim at the end of the reply.
• Respect Sif time conventions: week values must be Sunday dates; current week data may be delayed, so use latelyDay=7 for current-week-style checks.
• Do not expose or repeat MCP secret keys.

Field notes from real Sif MCP usage

Prefer MCP tools, but handle large outputs

Some Sif tool responses are large. If direct tool output is truncated or hard to parse, call the MCP HTTP endpoint with a small script and write a compact local JSON summary. Do not paste secrets into the final answer.

Recommended pattern: 1. Load the MCP URL and secret from .mcp.json if local config is available. 2. Call tools/call with the exact tool name and arguments. 3. Parse result.content[0].text as JSON. 4. Extract only business-relevant fields into a compact summary. 5. Read the compact summary before writing the final answer.

Use this when a full analysis needs many tools or when output exceeds the visible context.

Robust parsing notes

Sif fields may vary by tool or response mode:

• Variant count may appear as vaiantsNum or variantsNum.
• Variant attributes may be a list of dicts ([{value: "Buzz"}]) or strings (["Buzz"]).
• Traffic trend channel arrays may be arrays of objects with score, scoreRatio, and scoreChangeRatio, not plain numeric arrays.
• Keyword distribution numeric values may be strings; convert before arithmetic.
• Some tools return Chinese keys such as 官网验证; preserve verification links when useful.

When parsing, inspect keys first and write defensive extraction logic. If a field is absent, report the gap instead of fabricating a value.

Tool parameter pitfalls

• market_get_keyword_competition may reject the same keywords array style accepted by market_get_keyword_history; if it errors, continue with keyword history concentration (top3_click_share, top3_conversion_share), ASIN keyword signals, and root trend, and label competition-tool output as missing.
• ads_get_asin_campaign_contribution_overview requires an explicit start_date and end_date in some schemas. For a current snapshot, use a recent 30-day window.
• ads_get_asin_ad_window_feature_profile schema descriptions may mention window dates, but the observed schema can require granularity; check current tool schema if available.
• analyze_traffic_anomaly is for explicit diagnosis/root-cause requests, not normal full research.

Full-analysis minimum dataset

For a comprehensive ASIN report, collect at least: 1. ops_get_asin_sales_list and/or ops_get_asin_sales_trend. 2. ops_get_asin_traffic_trend. 3. ops_get_listing_traffic_overview. 4. ops_get_listing_traffic_structure or ops_get_listing_keyword_distribution. 5. market_get_asin_keyword_signals. 6. market_get_keyword_history for top 3-5 traffic keywords. 7. market_get_keyword_root_trend for the main exact/root terms. 8. Ads tools when ad share is material or the user asks for ads: ads_get_asin_ad_traffic_trend, ads_get_asin_ad_structure, ads_get_asin_ad_feature_profile, ads_get_asin_campaign_contribution_overview.

Do not call every drill-down tool by default. Drill down only if a decision depends on a campaign/ad group/keyword cause.

Interpretation lessons

• High sales plus high review count plus high natural share indicates a moat; do not recommend direct white-label entry without IP/compliance and differentiation checks.
• A product can have many campaigns while still being natural-led. Compare ad share against total traffic before calling it ad-driven.
• For parent/variant listings, separate parent-level demand from variant-level winners. Identify which styles or variants absorb most sales and traffic.
• Root coverage ratio changes strategy: low exact/root coverage means long-tail opportunity; high coverage means exact-term competition is the main battleground.
• Top3 click share near or above 0.6 means traffic is concentrated; below 0.3 means demand is more open.
• If a Sif competition tool fails, keyword history and ASIN keyword signals still provide usable concentration and ranking evidence, but confidence should be reduced for competition conclusions.

Final answer hygiene

• Never mention raw scripts, temporary file paths, or parsing errors unless the user asked about process details.
• Do mention data limitations that affect business confidence.
• Keep source links from Sif verification fields at the end when available.
• Use direct business language: 做 / 不做 / 小样测试, 加码 / 优化 / 止损, 防守词 / 进攻词 / 长尾词.

Sif research playbooks

1. Quick traffic report

Use when the user asks for ASIN traffic status or traffic composition.

Tools: 1. ops_get_asin_traffic_trend for total/natural/ad trend. 2. ops_get_listing_traffic_overview for natural vs ad share. 3. ops_get_listing_keyword_distribution for keyword coverage by channel or variant.

Output: trend direction, traffic dependency, weak channel, top follow-up checks.

2. Keyword reverse audit

Use when the user wants to know which keywords bring traffic to an ASIN or competitor.

Tools: 1. market_get_asin_keyword_signals for ranking, traffic contribution, and competitive position. 2. ops_get_listing_keyword_distribution for channel-level keyword counts or scores. 3. market_get_keyword_history for demand and concentration of the top 5-10 candidate terms. 4. market_get_keyword_competition for enterability of priority terms.

Output: main traffic terms, ad/natural split, defend terms, attack terms, missing terms.

3. Ad operations report

Use when the user asks whether advertising is healthy or which campaigns/ad groups matter.

Tools: 1. ads_get_asin_ad_traffic_trend for SP/SB/SBV trend. 2. ads_get_asin_ad_feature_profile or ads_get_asin_ad_window_feature_profile for concentration, rhythm, stability. 3. ads_get_asin_campaign_contribution_overview to rank campaigns. 4. Drill down only when needed: ads_get_campaign_traffic_trend, ads_get_campaign_contribution_breakdown, ads_get_ad_group_keyword_breakdown. 5. ads_get_asin_campaign_changes when performance change may be caused by budget/bid/status edits.

Output: dominant channel, concentration risk, waste/scale candidates, exact campaign/ad group next actions.

4. Traffic anomaly diagnosis

Use when the user asks why traffic/ranking/sales dropped or says traffic is abnormal.

Preferred tool:

• analyze_traffic_anomaly with asin, country, and optional time_type/time_value.

If manual corroboration is needed: 1. ops_get_asin_traffic_trend for when the anomaly started. 2. ops_get_listing_traffic_overview for natural vs ad split. 3. market_get_asin_keyword_signals for keyword rank/traffic shifts. 4. Ad tools only if ad share changed.

Output must include a cause tree or Mermaid diagram if tool output supports it, plus excluded hypotheses.

5. Competitor deep analysis

Use when the user provides competitor ASINs and asks how they grew or how to beat them.

Tools: 1. ops_get_asin_sales_list and/or ops_get_asin_sales_trend for sales, price, variant winners, seasonality. 2. ops_get_asin_traffic_trend and ops_get_listing_traffic_overview for growth stages and ad/natural dependency. 3. market_get_asin_keyword_signals for traffic source and ranking position. 4. market_get_keyword_history and market_get_keyword_competition for the top traffic terms. 5. Ad tools if the competitor relies heavily on paid traffic.

Output: growth path, core keywords, traffic engine, variant/price pattern, copyable tactics, hard-to-copy moat.

6. Keyword opportunity batch

Use when the user provides multiple keywords and wants prioritization.

Tools: 1. market_get_keyword_history for search volume, ABA rank, top3 concentration. 2. market_get_keyword_root_trend for exact-vs-root demand boundary. 3. market_get_keyword_demand for lifecycle and timing. 4. market_get_keyword_competition for enterability.

Score each keyword on demand, trend, concentration, competition, relevance, and expansion space. Label as 攻主词, 铺长尾, 观察, or 放弃.

7. Market validation

Use when deciding whether a product/market is worth doing.

Tools: 1. Keyword tools for demand size, trend, seasonality, concentration, and competition. 2. ops_get_asin_sales_list for competitor sales, price band, and variants. 3. Competitor ASIN keyword signals for traffic sources. 4. Traffic/ad tools when candidate ASINs show abnormal paid dependency.

Decision logic: demand stable/growing + non-monopoly + reachable price/profit band + clear keyword entry + tolerable review/brand moat = worth testing.

Output: 做 / 不做 / 小样测试, rationale, entry route, risk gates, minimum data needed before spend.

8. Full analysis

Use when the user asks for comprehensive research on an ASIN, product, or market. Also read field-notes.md first because full analysis commonly creates large Sif responses.

Minimum sequence: 1. Sales: ops_get_asin_sales_list for current sales, price, rating, review, variants; add ops_get_asin_sales_trend for seasonality. 2. Traffic: ops_get_asin_traffic_trend, ops_get_listing_traffic_overview, and either ops_get_listing_traffic_structure or ops_get_listing_keyword_distribution. 3. Reverse keywords: market_get_asin_keyword_signals; take the top 3-5 traffic keywords. 4. Market: market_get_keyword_history for those keywords; market_get_keyword_root_trend for main exact/root terms; use market_get_keyword_demand when timing/seasonality is material. 5. Ads: if ad share is material or the user asks for ads, call ads_get_asin_ad_traffic_trend, ads_get_asin_ad_structure, ads_get_asin_ad_feature_profile, and ads_get_asin_campaign_contribution_overview with an explicit recent 30-day date window. 6. Synthesis: decide whether the subject is a direct-entry opportunity, a moat/avoid case, an optimization case, or a long-tail opportunity.

If a tool errors or returns an unexpected schema, continue with available evidence, explicitly mark the missing layer, and avoid overstating confidence.

Sif MCP tool map

MCP server

Configured as sif-mcp using HTTP MCP. Use the connected MCP tools exposed by the client; do not hard-code the server URL or secret key in answers.

Common parameters

Check current tool schemas when possible; Sif schemas can differ from prose descriptions. If a call fails, inspect the error, adjust only that tool's parameters, and continue with other evidence.

• country: marketplace code. Supported examples: US, UK, DE, CA, JP, FR, ES, IT, MX, AU, AE, BR, SA.
• timePieceType: latelyDay, week, or month.
• timePieceValue: for latelyDay use 7 or 30; for week use the Sunday date yyyy-MM-dd; for month use the first day of month yyyy-MM-dd.
• granularity: usually week or month; use day only when the tool supports it and the user needs short-window detail.
• Pagination: use pageSize up to the tool max, usually 20-200. Fetch more pages only when needed for the decision.

Market and keyword tools

• market_get_keyword_history: exact keyword search volume, ABA rank, Top3 click/conversion concentration. Best for current demand size and concentration.
• market_get_keyword_root_trend: exact keyword vs root-level total demand. Best for market boundary and long-tail opportunity.
• market_get_keyword_demand: lifecycle, demand trend, and timing. Best for entry timing and seasonality.
• market_get_keyword_competition: competitive structure and enterability. Best for deciding whether a keyword can be attacked.
• market_get_asin_keyword_signals: ASIN keyword ranking, traffic contribution, and competitive position. Best for reverse-engineering competitors.
• ops_get_listing_keyword_distribution: listing/variant keyword coverage by natural and ad channels. Best for coverage gaps and traffic source split.

Operations tools

• ops_get_asin_traffic_trend: ASIN/listing traffic time series by natural/ad/SP/SB/SBV. First stop for trend questions.
• ops_get_asin_traffic_trend_detail: keyword-level traffic detail in a specified time window. Use after trend identifies the window.
• ops_get_listing_traffic_overview: listing natural vs ad share and ad channel breakdown.
• ops_get_listing_traffic_structure: variant-level traffic structure by natural/ad channels.
• ops_get_asin_sales_trend: monthly sales trend by ASIN/variant/dimension.
• ops_get_asin_sales_list: sales, price, attributes, and short trend for one or more ASINs.

Advertising tools

• ads_get_asin_ad_traffic_trend: ASIN SP/SB/SBV exposure trend and dominant channel.
• ads_get_asin_ad_structure: campaigns and ad groups under the ASIN.
• ads_get_asin_ad_feature_profile: ASIN ad structure, rhythm, complexity, stability.
• ads_get_asin_ad_historical_feature_profile: historical change in ad style.
• ads_get_asin_ad_window_feature_profile: ad profile for a chosen date window.
• ads_get_asin_campaign_contribution_overview: campaign contribution ranking in a date window.
• ads_get_asin_campaign_changes: bid/budget/status changes; use for suspected operational causes.
• ads_get_campaign_structure: ad groups in a campaign.
• ads_get_campaign_traffic_trend: campaign lifecycle trend and anomaly weeks.
• ads_get_campaign_contribution_breakdown: contribution by ad group or keyword within campaign.
• ads_get_ad_group_traffic_trend: ad group trend.
• ads_get_ad_group_keyword_breakdown: keywords and display ASINs inside an ad group for a week.

Scenario tool

• analyze_traffic_anomaly: end-to-end ASIN traffic drop root-cause diagnosis. Use for explicit anomaly diagnosis; not for raw data display.

Interpretation thresholds

• Top3 click share > 0.6: traffic is highly concentrated; new entrants face high traffic acquisition difficulty.
• Top3 click share 0.3-0.6: moderate concentration; entry depends on differentiation and keyword route.
• Top3 click share < 0.3: distributed demand; more room for new entrants.
• Root coverage ratio > 0.8: demand is concentrated in the exact term.
• Root coverage ratio 0.4-0.8: mix exact and long-tail layout.
• Root coverage ratio < 0.4: demand is dispersed; exact term alone underestimates market size.

Treat thresholds as heuristics. Confirm with competition, sales, price, review moat, and product relevance before recommending execution.