The Semrush MCP server connects Semrush’s knowledge on to AI assistants like Claude and ChatGPT, so you’ll be able to run actual search engine optimization analysis in plain language. The Model Context Protocol (MCP) is an open normal created by Anthropic that offers AI fashions a common method to hook up with exterior knowledge sources, information, and instruments.
If you happen to’ve heard of MCP however have not put it to work but, it’s laborious to know what to ask.
This information offers you 16 Semrush MCP use circumstances with copy-paste prompts, grouped by workflow: key phrase technique, aggressive intelligence, content material optimization, and diagnostics.
How one can arrange the Semrush MCP
Setting up the Semrush MCP takes 4 steps:
- Examine your plan. MCP entry comes with Semrush One Starter, Semrush One Professional+, search engine optimization Basic Professional, and search engine optimization Basic Guru, every together with 50,00Zero API models. Visitors & Market experiences want a separate Developments API subscription, which I will flag when it comes up.
- Join from inside your AI consumer. In Claude, go to “Settings”, then “Connectors.” Click on “Add,” then “Browse Connectors.” Seek for Semrush MCP, and approve the permissions. In ChatGPT, go to Settings, then Apps, discover Semrush, and click on Join. Each use OAuth, so there isn’t any key to stick.
- Use the endpoint for different purchasers. Cursor, VS Code, Gemini, Perplexity, and customized brokers connect with https://mcp.semrush.com/v2/mcp with an API key within the Authorization header. The developer docs have the config for each. If you happen to’re working in a terminal, Claude Code with Semrush is identical concept with extra automation on high.
- Verify the connection. Ask one thing low cost, like “What is the Semrush Rank for my area within the US database?” If a quantity comes again, you are reside. If there’s an error, ask the AI to stroll you thru fixing it.
Each immediate under is within the Semrush MCP prompt library, the place every use case is a workflow of three or 4 chained prompts.
I’ve featured the primary immediate in every workflow, since that is the one which pulls the information, and linked the total workflow so you’ll be able to run the follow-ups. To make use of one, paste it, swap your individual area, nation, and key phrases into the {brackets}, run it, and browse the output earlier than you act on it.
Search demand & key phrase technique
These search demand and key phrase technique prompts uncover the place the demand in your area of interest really sits, which key phrases rivals personal that you do not, and which gaps deserve effort first.
Spot shifts in search demand
This immediate maps your area of interest’s greatest key phrase clusters by mixed quantity earlier than the full workflow layers rising, declining, and SERP-opportunity views on high. Use it while you’re planning 1 / 4 and have to know the place demand lives earlier than deciding what to construct.
Utilizing Semrush key phrase knowledge for {nation}:
Establish the high 8 key phrase clusters for "{area of interest}" by mixed month-to-month search quantity.
Return ONE desk:
Columns:
* cluster_name
* combined_monthly_volume
* example_keywords (up to 5)
Restrict: 8 clusters precisely.
You get a desk of eight clusters ranked by mixed quantity. Subsequent, run the workflow’s development prompts to see which clusters are rising.
Flip key phrase gaps into roadmaps
The immediate under finds key phrases rivals rank for that you do not, plus those the place you rank far behind. The full workflow then clusters them and plans pages. Attain for it when you already know site visitors goes to rivals however not by means of which doorways.
If {competitor-domains} are offered, use them straight (up to 5). If not, first discover {your-domain.com}'s high natural search rivals (restrict 5); exclude domains with Competitor Relevance = 0.00 and natural site visitors > 10M (e.g., YouTube, Reddit, Wikipedia).
Use Semrush knowledge for {nation} to analyze {your-domain.com} in opposition to its high natural rivals.
Discover and prioritize two alternative varieties:
* Lacking key phrases: rivals rank, however {your-domain.com} does not.
* Weak shared key phrases: each rank, however {your-domain.com} ranks a lot decrease than the strongest competitor.
Prioritize low-hanging fruit that look actionable by means of content material or on-page enhancements.
Return ONE desk (up to 50 rows):
Columns:
* key phrase
* opportunity_type
* monthly_volume
* intent
* top_competitor_domain
* competitor_rank
* your_rank
* rank_gap
* recommended_action
* rationale
If direct hole evaluation is unavailable, approximate utilizing competitor key phrase overlap plus ranking-gap comparability, and clarify the methodology briefly.
You get a 50-row desk splitting gaps into “lacking” and “weak shared” with a advisable motion per row. The workflow’s subsequent prompts cluster the record into themes.
Prioritize gaps by demand and intent
This immediate estimates the search intent combine inside every hole cluster so you’ll be able to sequence them by enterprise worth moderately than uncooked quantity. The full workflow then carries it by means of competitor problem checks right into a six-week dash plan. Run it as a follow-up as soon as a spot record exists.
Utilizing Semrush key phrase knowledge for {nation}:
For the 8 hole clusters, estimate intent distribution.
If hole clusters already exist in this dialog, use them.
If no hole clusters are out there, first establish key phrase gaps for {your-domain.com} in opposition to up to 5 competitor domains, then cluster them into precisely 8 themes.
Return ONE desk:
Columns:
* cluster_name
* informational_share_pct (est)
* commercial_share_pct (est)
* transactional_share_pct (est)
* top_intent_keywords (up to 5)
Restrict: 8 rows.
If intent labels are unavailable, infer from SERP/web page varieties and label as estimate.
The output is an eight-row desk of estimated intent shares per cluster. Be sure to spot-check a SERP or two earlier than trusting the roadmap, since they’re model-generated estimates.
Aggressive intelligence
These aggressive intelligence prompts establish who you really compete with in search, how the site visitors splits, who’s rising, and watch them with out dwelling in dashboards.
Establish your true search rivals
The immediate under ranks the domains sharing your key phrases by Semrush’s Competitor Relevance rating moderately than by who you assume you compete with; the full workflow maps overlap clusters and SERP function wins subsequent.
Utilizing Semrush knowledge for {nation}:
Establish the high 10 natural rivals of {your-domain.com}, excluding high-traffic generic domains (Competitor Relevance = 0.00 or natural site visitors > 10M, e.g., YouTube, Reddit, Wikipedia).
Return ONE desk:
Columns:
* competitor_domain
* estimated_organic_traffic
* ranking_keywords
* keyword_overlap_with_{your-domain.com}
* overlap_pct (if out there)
Restrict: high 10 rivals.
If overlap metrics are not out there, return best-effort and label lacking fields as N/A.
The output is a 10-row desk with site visitors, key phrase counts, and overlap per area. Feed these names into the following 4 prompts.
Prioritize strengths, gaps, and assaults
This immediate finds the clusters the place you are sturdy and rivals are weak, so you already know what to defend earlier than selecting what to chase; the full workflow ends in a defend-versus-attack map.
Utilizing Semrush knowledge for {nation}:
Establish key phrase clusters the place {your-domain.com} has comparatively sturdy visibility however the high 5 rivals have weaker presence.
If a earlier competitor focus or cluster evaluation exists in this dialog, use it as a beginning level. Give particular consideration to clusters beforehand recognized as open, untargeted, weakly coated by rivals, or sturdy alternatives for {your-domain.com}.
If no prior evaluation is out there, establish the clusters straight from Semrush key phrase and competitor knowledge.
Return ONE desk:
Columns:
* unique_cluster
* why_unique (1 sentence)
* example_keywords (up to 5)
* suggested_defense_action
Restrict: 8 clusters.
If "relative visibility" is unavailable, infer utilizing rating key phrase protection and label as inferred.
You stand up to eight clusters you lead, every with a protection motion. The workflow’s subsequent prompts then present the place rivals outrank you.
Dimension site visitors share throughout rivals
Use the immediate under to tug every area’s site visitors and engagement and compute market share. Observe that it calls Traffic Analytics, which wants a Developments API subscription; the full workflow continues into channel and geography splits.
For every competitor area ({competitor-domains}) in {nation}, use Semrush Visitors Analytics to retrieve every area's general site visitors abstract (visits, distinctive guests, engagement). It accepts a number of domains per request — move all domains collectively in a single name. If {competitor-domains} are not offered, use competitor domains recognized earlier in this use case as high natural/search rivals, market rivals, or strongest keyword-overlap rivals.
If each {competitor-domains} and beforehand recognized rivals are out there, mix them, take away duplicates, and restrict to the 5 most related competitor domains.
Request columns:
* goal
* rank
* visits
* customers
* pages_per_visit
* bounce_rate
* time_on_site
Construct a single comparability desk:
| Area | Rank | Visits | Distinctive Guests | Pages/Go to | Bounce Charge | Avg Period (s) | Visitors Share % |
Visitors Share % = every area's visits / sum of all visits (calculate after all knowledge is returned).
After the desk, derive:
* total_market_traffic: sum of all visits
* market_leader: area with highest visits
* traffic_concentration: mixed site visitors share % of the high 2 domains
If knowledge unavailable for a area, label as low_data. Return all knowledge in a single response.
Again comes a share-of-market desk with whole market site visitors, the market chief, and top-two focus; on a plan with out the Developments API, the MCP experiences the hole and falls again to natural estimates, so low_data engagement columns imply your plan, not a damaged immediate.
Discover rivals gaining natural site visitors
To seek out rivals gaining natural site visitors, run the immediate under. It pulls 12 months of site visitors historical past per competitor and ranks them by absolute progress; the full workflow turns the winners’ patterns into playbooks.
Utilizing Semrush knowledge for {nation} over the final 12 months, establish the high 10 rivals of {your-domain.com} by natural site visitors progress.
Course of:
1. Discover the high natural rivals of {your-domain.com}
2. Choose high 10 by key phrase overlap or aggressive relevance
Exclude domains with Competitor Relevance = 0.00 and Natural Visitors > 10M — these are mega-platforms (YouTube, Reddit, Fb), not area of interest rivals.
3. For every competitor, pull its natural site visitors development over time (one name per area)
4. Extract: traffic_12m_ago, traffic_now
5. Compute:
- traffic_change_abs = traffic_now − traffic_12m_ago
- traffic_change_pct = (traffic_change_abs / traffic_12m_ago) × 100
Flag any competitor the place traffic_12m_ago < 100 as low_data — % progress from a tiny base is deceptive.
6. Decide top_growth_cluster driving site visitors progress
Return ONE desk:
Columns:
* competitor_domain
* traffic_change_abs
* traffic_change_pct
* top_growth_cluster
Restrict: high 10 by traffic_change_abs.
Full all sequential calls earlier than returning the closing desk.
Use closest out there dates if 12-month knowledge is incomplete.
You get a progress leaderboard with the cluster driving every acquire. The low_data flag issues, as a result of proportion progress from a tiny base will in any other case high the desk.
Monitor competitor visibility shifts
To trace competitor visibility shifts, set the watch record first with the immediate under. It builds a monitoring desk of your 10 most related rivals and the cluster every competes on. The full workflow then establishes the baseline you measure shifts in opposition to.
Utilizing Semrush knowledge for {nation}:
Establish the high 10 natural rivals of {your-domain.com} to monitor. Kind by competitor relevance (Cr) descending.
Return ONE desk:
Columns:
* competitor_domain
* estimated_organic_traffic
* keyword_overlap (if out there)
* primary_competing_cluster (1)
Restrict: 10 rivals.
Out comes a compact watch record. Regenerate the record month-to-month and hand it to the alerts immediate under.
Construct alerts and response performs
Use this immediate to get alerts when a competitor good points rankings otherwise you lose them. Do not forget that the MCP writes the principles however cannot create alerts. Implement the output inside Semrush, for instance as Position Tracking campaigns; the full workflow provides response playbooks.
Create a competitor monitoring alert ruleset for {your-domain.com} in {nation}.
These alerts are designed to detect when monitored rivals make significant strikes and when {your-domain.com} loses floor. Do NOT generate alerts for {your-domain.com} good points — the function is early warning, not reporting success.
Cowl two sign classes:
* competitor_gain: a monitored competitor good points natural site visitors, rankings, or visibility above threshold in clusters overlapping with {your-domain.com}
* own_loss: {your-domain.com} drops in rankings, site visitors share, or key phrase visibility in a monitored cluster
Base metric indicators on natural site visitors development over time and on place modifications in shared key phrases (natural search outcomes).
Return ONE desk:
Columns:
* alert_name
* signal_type (competitor_gain / own_loss)
* metric
* threshold
* cadence
* action_owner_role
* what_to_investigate
Embrace at least 8 alert guidelines — minimal 5 of sort competitor_gain, minimal 2 of sort own_loss.
You get a guidelines desk with thresholds, cadences, and homeowners. If you happen to run this in the identical dialog as the sooner aggressive prompts, the AI will calibrate in opposition to actual baselines within the dialog and each threshold will carry each a proportion and an absolute ground as a substitute of a generic quantity. Backtest it in opposition to a previous yr’s knowledge earlier than you settle for the output.
Content material creation & optimization
These content material creation and optimization prompts discover the pages shedding site visitors, baseline them for refreshes, map demand round your product, and switch competitor gaps right into a publish plan.
Prioritize declining pages for restoration
To seek out and prioritize declining pages for restoration, run the immediate under. It compares your pages in opposition to a six-month-old snapshot and traces every drop to a key phrase place change. The full workflow then scores restoration potential.
Utilizing Semrush knowledge for {nation}:
Establish pages on {your-domain.com} that have misplaced the most natural site visitors over the previous 6 months.
Step 1 — Use Semrush Natural Analysis to get {your-domain.com}'s pages by natural site visitors, sorted by site visitors ascending. Exclude the homepage, pagination pages, and tag/class pages.
Step 2 — For every of the high 20 pages by lowest present site visitors: use Semrush to pull every web page's rating historical past for the previous 6 months for the high key phrase per web page.
Return ONE desk:
Columns:
* page_url
* current_monthly_traffic
* traffic_6mo_ago (estimated)
* traffic_change (%)
* top_keyword
* current_position
* position_6mo_ago
Restrict: 20 pages.
Kind by traffic_change ascending (largest drop first). If historic knowledge is unavailable for a web page, mark as low_data.
The output is a decline desk sorted by greatest drop, with the key phrase behind every. Learn the AI’s interpretation notes too, as it could catch errors you may in any other case miss.
Create briefs and drafts for refreshes
This immediate fetches your reside pages and combines their present content material along with your Semrush key phrase baselines; the full workflow then finds key phrase targets and drafts a refreshed article.
For every web page chosen for content material refresh, extract the present content material and search engine optimization efficiency baseline for {your-domain.com} in {nation}.
Pages to analyze:
* If {paste URLs right here} are offered, embrace these pages.
* If a refresh backlog already exists in this dialog, additionally embrace all pages marked `full_rewrite` or `targeted_update`.
* If each sources are out there, mix them and take away duplicate URLs.
For every web page:
1. Fetch the present web page content material.
2. Pull the web page's present key phrases, positions, and estimated site visitors ({nation}).
Return ONE desk:
Columns:
* page_url
* supply (provided_url / refresh_backlog / each)
* word_count
* h1
* meta_description (first 160 chars)
* top_3_keywords (key phrase: place)
* estimated_monthly_traffic
* content_issues (skinny / outdated / missing_keywords / none)
Content material difficulty guidelines:
* skinny = word_count < 800
* outdated = content material references particular years prior to {current_year} - 2
* missing_keywords = web page ranks for fewer than 5 key phrases
* none = no apparent points detected
If a number of points apply, record all separated by commas.
Full all calls earlier than returning the closing desk.
You get one row per web page with difficulty flags connected. That baseline carries straight into the workflow’s brief-writing step.
Flip product demand into web page blueprints
This immediate maps search demand on your product throughout 5 intent clusters and reads the SERP for every. The full workflow then converts the map into web page constructions, and its sibling workflow prepares pages for newer surfaces. This might develop into extra essential as agentic commerce and the Universal Commerce Protocol route product discovery by means of AI brokers.
Utilizing Semrush key phrase knowledge for {nation}:
Map the search demand panorama for the product in class {area of interest}.
Step 1 — Establish 3–5 core intent clusters:
* product title / branded searches
* class searches (generic)
* use-case searches ("finest {area of interest} for [job]")
* comparability searches ("{area of interest} vs", "{area of interest} alternate options")
* transactional modifiers ("purchase {area of interest}", "{area of interest} worth")
Step 2 — For every cluster, run key phrase analysis and extract the high 2–3 consultant key phrases by way of a bulk key phrase lookup (a number of key phrases in one request).
Step 3 — For every key phrase return: search quantity, key phrase problem, intent (business / transactional / informational), and dominant SERP web page sort. Look at the natural search outcomes (together with SERP options) to establish SERP web page varieties and energetic SERP options if the bulk lookup does not return SERP-feature knowledge.
Return ONE desk:
Columns:
* intent_cluster
* example_keyword
* monthly_volume
* kd
* intent
* dominant_serp_page_type
* alternative (excessive/med/low)
Alternative rule:
* excessive = quantity ≥ 500 AND kd ≤ 65 AND intent = business or transactional
* med = quantity 100–499 OR kd 66–80
* low = quantity < 100 OR kd > 80
Restrict: 12 rows.
Full all key phrase calls earlier than returning the desk.
If quantity knowledge is unavailable for a key phrase, mark as low_data.
Additionally return: ONE sentence — main key phrase suggestion for the product web page title tag.
You get a 12-row demand map plus a title tag suggestion.
Flip gaps into publish-ready plans
This immediate takes your high content material hole clusters, checks whether or not you have already got a web page for every, and classifies each cluster as a refresh or a create; the full workflow then specs the briefs.
Utilizing the high 3 prioritized content material hole clusters and their existing_coverage knowledge:
If prioritized content material gaps already exist in this dialog, use the high 3 clusters from that output.
If no prioritized content material gaps are out there, first establish and prioritize content material gaps for {your-domain.com} in {nation}:
* If competitor domains are offered, use them straight, up to 5 domains.
* If no rivals are offered, establish high natural rivals for {your-domain.com} utilizing Semrush knowledge for {nation}.
* Establish content material matter clusters rivals cowl the place {your-domain.com} has low or no visibility.
* Examine whether or not {your-domain.com} seems in the high 50 outcomes for every cluster's main key phrase.
* Prioritize clusters by demand, key phrase problem, present protection, and enterprise match.
* Choose the high 3 clusters.
Classify every cluster:
* refresh: existing_coverage = sure — {your-domain.com} already has a web page on this matter
* create: existing_coverage = no — no present web page; construct from scratch
Return ONE desk:
Columns:
* cluster_name
* action_type (refresh / create)
* existing_page_url (from protection examine, or n/a)
* priority_rank
* why_prioritized (1 sentence)
Restrict: 3 rows.
The output is a three-row plan, every cluster tagged “refresh” or “create” with a one-sentence case. Decide the competitor set fastidiously, since one dangerous area can pollute the entire hole pull.
Diagnostics & reporting
These diagnostics and reporting prompts discover fast wins, bundle efficiency for management, and separate site-specific site visitors losses from market-wide ones.
Floor fast-moving search engine optimization alternatives
This immediate pulls your key phrases rating in positions 5 to 20 and tiers them by a quick-win rule that mixes place, quantity, and problem. The full workflow converts the winners right into a dash.
Utilizing Semrush rating knowledge for {nation}:
Discover the high 50 key phrases for {your-domain.com} at the moment rating positions 5–20.
Return ONE desk:
Columns:
* key phrase
* current_position
* monthly_volume
* keyword_difficulty
* landing_page
* intent (if out there)
* quick_win_signal (excessive/med/low)
Fast win sign rule:
* excessive: place 5–10 AND monthly_volume ≥ 500 AND keyword_difficulty ≤ 60
* med: place 11–20 OR monthly_volume 100–499 OR keyword_difficulty 61–75
* low: monthly_volume < 100 OR keyword_difficulty > 75
Kind: excessive first, then by monthly_volume descending.
Restrict: 50 rows.
Out comes a 50-row tiered record with touchdown pages connected; on my run the “excessive” tier was largely low-value dictionary queries whereas the true wins sat in “med,” so re-rank by enterprise relevance earlier than executing it. The MCP finds the rows; deciding which rows matter remains to be your job.
Flip search engine optimization insights into board-ready choices
Run the immediate under to compress your aggressive place and site visitors development into three tables: a competitor snapshot, 5 insights with cited metrics, and 5 choices with urgency rankings. It is the second half of the executive reporting workflow.
Utilizing Semrush knowledge for {your-domain.com} in {nation} (final 30 days):
1. Discover {your-domain.com}'s high 5 natural search rivals, sorted by aggressive relevance, highest first. Columns: area, organic_traffic, competitor_relevance.
2. Pull {your-domain.com}'s natural site visitors development over time — retrieve final 3 months of knowledge.
Construct Desk A — Competitor snapshot (5 rows):
Columns:
* competitor_domain
* est_organic_traffic
* competitor_relevance
* traffic_vs_your_domain (% of your site visitors)
Construct Desk B — 5 strategic insights:
Columns:
* perception
* data_signal (cite particular metric)
* action_point (1 sentence)
Construct Desk C — 5 choices management ought to make:
Columns:
* determination
* reasoning (1 sentence)
* urgency (excessive/med/low)
If a efficiency snapshot, dangers, or alternatives already exist in this dialog: incorporate the efficiency snapshot, dangers, and alternatives from these steps into Tables B and C.
Restrict: 5 rows every for Tables B and C.
Label estimates as "estimate". Full all calls earlier than returning the tables.
You get a one-pager you’ll be able to carry to a management assembly.
Hint natural site visitors losses to root causes
This immediate tells you whether or not the entire market dropped or simply you, by evaluating your decline window in opposition to 5 rivals. The full workflow then classifies the scope of a site-specific drop.
Earlier than diagnosing site-level causes, decide whether or not the natural site visitors drop for {your-domain.com} in {nation} displays an industry-wide occasion.
Step 1 — Sensor reference: notice the approximate drop window. Examine Semrush Sensor (semrush.com/sensor) manually for that interval — SERP volatility above 3.5 indicators a confirmed or suspected algorithm replace.
Step 2 — Market comparability: establish 5 direct natural rivals of {your-domain.com}. For every competitor area, use Semrush Visitors Analytics to retrieve visits for the drop window and an equal prior interval. Name one area at a time.
Return ONE desk:
Columns:
* area
* visits_pre_drop
* visits_during_drop
* traffic_change_pct
* also_dropped (sure/no)
After the desk:
* industry_wide_signal: true if 3 or extra competitor domains additionally present a detrimental traffic_change_pct throughout the identical interval
* conclusion: if industry_wide_signal is true → "Possible algorithm replace or market-wide SERP occasion — correlate with Semrush Sensor earlier than persevering with"; if false → "Drop seems site-specific — proceed to scope classification"
Return all knowledge in a single response. If Visitors Analytics knowledge is unavailable for a area, use the Area Overview instrument or the Natural Analysis for the natural site visitors development knowledge. Full all calls earlier than returning the desk.
The output is a comparability desk plus an express verdict. Observe that Semrush Sensor is not uncovered by means of the MCP, so the volatility examine remains to be handbook.
Suggestions for getting extra out of the Semrush MCP
Getting extra out of the Semrush MCP largely comes right down to prompting it like an analyst:
- Be particular. Identify the area, market, and timeframe so the MCP makes one exact request as a substitute of a number of imprecise ones.
- Give it one job at a time. Chain use circumstances throughout messages; the library’s workflows are constructed this manner, and later prompts reuse earlier outputs from the identical dialog.
- Set the database and machine wherever the information is dependent upon it. Each library immediate carries a {nation} placeholder because of this.
- Ask the AI to point out the numbers it pulled so you’ll be able to sanity-check its judgment calls earlier than appearing on them.
- Watch your API models. Each MCP request spends Semrush API models, and unit prices scale with how a lot knowledge a question requests, so “high 50 key phrases within the US” prices lower than “all key phrases globally.” Specificity is effectivity.
- Adapt the prompts. Regulate the enter in {brackets}. Change thresholds, add a business-value tiebreaker to the quick-wins rule, or match alerts to your web site’s actual site visitors numbers.
My take a look at web site had a 36% site visitors drop, a competitor at parity, and a duties web page shedding to a single legislation agency article. Each a kind of info was sitting in Semrush’s knowledge ready for somebody to search out it. Decide a use case from the Semrush MCP prompt library and discover the information hiding underneath your nostril.
