A Semrush MCP where the row limit is the spending control
Connect one Semrush API key and 28 nodes land on the canvas: domain overview and organic and paid footprints, competitor overlap, shopping ads, keyword research, URL and subdomain reports, the v4 backlinks family and project management. Semrush bills most reports PER LINE RETURNED, so every node names its unit cost, the balance check is free, and the connection test spends nothing.
What the Semrush MCP connector does
The Semrush connector for Mattering Automations is 28 nodes across both halves of the Semrush API, authenticated with one key. Eleven typed nodes cover the classic Analytics API, which is a single address whose meaning is chosen by a report type, and each typed node offers only its own family of report types as a validated choice: domain overview, domain organic search, domain paid search, domain competitors, domain shopping ads, keyword overview, keyword ideas, keyword SERP and advertisers, URL report, subdomain report and market leaders and movers. Semrush: Run Any Report is the escape hatch that accepts all 39 documented report types plus anything Semrush ships later. Nine nodes cover the v4 backlinks family, one covers v4 keyword metrics, five cover projects, and one reads the unit balance.
Money first, because on this API the row limit is a price and not a preference. Semrush meters in API units, and most Analytics reports bill PER LINE RETURNED. One domain organic call for 1,000 rows at 10 units per line is 10,000 units. The unit price differs per report and each node states its own: the domain competitor family reaches 80 units per line on the widest report, several keyword reports are 40, and pinning a historical month multiplies the price by five on the reports that publish a historical rate. A few endpoints bill flat per request instead, notably the backlinks overview at 45 units and the v4 keyword metrics call at 20. Semrush: Authority Score Distribution is the cheap one at 1 unit per line, which makes it a sensible way to size a link profile before enumerating it. Semrush: API Units Balance costs nothing at all.
Every node speaks MCP, the open protocol for connecting AI assistants to tools, so Claude, ChatGPT, and any MCP-capable assistant can use them, alongside Matter, the AI assistant built into the platform. An agent can read Semrush: API Units Balance for free before it does anything, size a link profile cheaply, and only then spend on the full pull. The same nodes run unattended on a schedule, which is where the balance check earns its place: on a per-line API the guardrail has to run when nobody is watching.
Two properties of the underlying API shape how this connector behaves, and both are stated on the nodes rather than left to be discovered. Semrush answers the classic reports as semicolon-delimited CSV rather than JSON, so the connector parses it into a header list and one object per line keyed by the vendor's own two-letter column codes, capped at 10,000 rows with a truncation flag rather than silently. And Semrush reports failures as a plain-text error line, frequently alongside an HTTP 200, so treating a 2xx as success would surface a wrong key, a disabled API or an exhausted unit balance as an empty report. Every node inspects the body instead. The nothing-found case is handled as its own thing, so a workflow can tell an empty answer apart from a broken credential.
Semrush ships its own MCP server and this page will not pretend otherwise. Semrush documents a hosted endpoint with OAuth as the default authentication approach and API key authentication available, exposing discovery tools for domain overview, organic research, keyword research, competitor research, backlinks research, audience research, traffic overview, paid search research, shopping research, position tracking, site audit and projects, plus a report schema tool and a generic report runner. Notably, Semrush gates it on a different set of subscriptions than the Standard API this connector uses. If you want to ask Semrush questions in a chat window, use theirs. This connector is 28 typed nodes that run on a schedule or a trigger with nobody present, chain with Google Search Console and every other connector in one workflow, hold the key once in a shared workspace store, and carry their per-line unit price on the node so a workflow can be reviewed for spend before it runs.
One connection, three surfaces
- Visual automations. Drag Semrush nodes onto the canvas and wire them to triggers, conditions, and other apps.
- AI agents, yours or ours. These nodes speak MCP, the open protocol, so Claude, ChatGPT, and any MCP-capable assistant can use them, and Matter, the agent built into the platform, uses them natively.
- The whole platform. Combine with native tasks, projects, and other connectors, no middleware in between.
Every Semrush node, ready to drop on the canvas
28 nodes, generated from the connector itself so this list always matches what ships. Expand a category to browse.
Read and query nodes 25
Semrush: API Units Balance
How many API units the account has left. Costs NOTHING to call: it is the only node here that spends no units, which makes it the right thing to put in front of an expensive branch: read the balance, compare it against what the next pull will cost, and stop the workflow rather than have Semrush answer ERROR 132 halfway through. The endpoint answers a bare integer as plain text, not JSON, so this connector parses it into api_units and keeps the raw body beside it. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Authority Score Distribution
How a target's referring domains are distributed across Authority Score bands: one row per score bucket with how many domains sit in it. A profile weighted to the low end is a quality problem the raw counts hide. SPENDS API UNITS: 1 unit per LINE, which makes this by far the cheapest backlinks report. The domain counts are a distribution of a single snapshot, so they may be summed WITHIN one response and never across responses. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Backlink Anchors
The anchor texts pointing at a target, with how many backlinks and how many distinct domains use each one. The anchor distribution is what an over-optimisation audit reads. SPENDS API UNITS: 40 units per LINE. Rows are ranked and truncated at Row Limit, so two periods' top-anchor lists must never be added together. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Backlink Competitors
Domains with a similar backlink profile to yours, scored by similarity and by how many referring domains you share. The starting point for a link-gap campaign. Note this endpoint takes a plain domain and has NO scope parameter, unlike the rest of the backlinks family. SPENDS API UNITS: 40 units per LINE. Similarity is a point-in-time score, not a count. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Backlinks
The individual backlinks pointing at a target: source URL and title, target URL, anchor text, the source page's and domain's Authority Scores, when the link was first and last seen, and whether it is nofollow, sponsored or UGC. SPENDS API UNITS: 45 units per LINE, which makes this the most expensive backlinks node, and a 1,000-link pull is 45,000 units. Size the profile with Backlinks Overview first (45 units flat), then filter server-side. Authority Scores are point-in-time readings and must not be summed or averaged across pulls. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Backlinks Historical Summary
The link profile month by month, so you can chart whether a site's backlinks and referring domains are growing: one row per month with Authority Score, backlink count, referring domain count and follow count. This is the ONLY node in the connector that takes a real start and end date, which is why it is the only one carrying a window declaration. SPENDS API UNITS: 40 units per line, so a 24-month pull is 960 units. Each row is a SNAPSHOT of the cumulative profile at that month, not that month's activity, so the counts must be read as levels and never added together. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Backlinks Overview
One row summarising a target's whole link profile: total backlinks and referring domains, follow vs nofollow vs sponsored vs UGC, Authority Score, referring IPs and class-C networks, link types, and how many links were gained and lost. SPENDS API UNITS: 45 units per REQUEST (not per line), so this is the cheap way to size a profile before pulling the links themselves. Authority Score is a point-in-time 0-100 reading and the counts are cumulative totals as of now, so nothing here may be added across pulls. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Domain Competitors
Who you are actually competing with. domain_organic_organic ranks the domains sharing your organic keywords by relevance, domain_adwords_adwords does the same for paid, and domain_domains compares up to five named domains keyword by keyword so you can see who ranks where for each phrase. This is the most expensive family in the connector: domain_domains is 80 units per LINE. SPENDS API UNITS, per line returned, so Row Limit is a spending control and not a convenience: domain_organic_organic costs 40 API units per line, or 200 per line when Historical Month is set; domain_adwords_adwords costs 40 API units per line, or 200 per line when Historical Month is set; domain_domains costs 80 API units per line, or 400 per line when Historical Month is set. Traffic and search-volume figures here are Semrush's MODELLED estimates derived from its own SERP crawl and clickstream panel, not measured visits to the site. Treat them as directional, and never present them beside real analytics numbers as though they were the same kind of quantity. Positions, visibility and authority scores are POINT-IN-TIME readings, not counts: they cannot be added across periods, and an unweighted average of them is also wrong. That is why these nodes declare no aggregatable measures. Output shape: Semrush answers this endpoint with semicolon-delimited CSV, not JSON, so this connector parses it. data.columns is the header row, data.rows is one object per line keyed by those column codes, and data.row_count says how many came back. Values are the vendor's own text, uncast, because one report can mix decimals, packed date stamps and comma-joined trend series in a single table. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Domain Organic Search
Every organic keyword a domain ranks for in one database, with position, estimated traffic share, CPC, difficulty, intents and SERP features. domain_organic_pages rolls the same data up to the pages that earn it and domain_organic_subdomains rolls it up to subdomains. This is the workhorse report and the expensive one at volume. SPENDS API UNITS, per line returned, so Row Limit is a spending control and not a convenience: domain_organic costs 10 API units per line, or 50 per line when Historical Month is set. domain_organic_pages and domain_organic_subdomains are metered per line on the same basis. Traffic and search-volume figures here are Semrush's MODELLED estimates derived from its own SERP crawl and clickstream panel, not measured visits to the site. Treat them as directional, and never present them beside real analytics numbers as though they were the same kind of quantity. Positions, visibility and authority scores are POINT-IN-TIME readings, not counts: they cannot be added across periods, and an unweighted average of them is also wrong. That is why these nodes declare no aggregatable measures. Output shape: Semrush answers this endpoint with semicolon-delimited CSV, not JSON, so this connector parses it. data.columns is the header row, data.rows is one object per line keyed by those column codes, and data.row_count says how many came back. Values are the vendor's own text, uncast, because one report can mix decimals, packed date stamps and comma-joined trend series in a single table. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Domain Overview
The headline numbers for a domain: Semrush Rank, organic and paid keyword counts, estimated traffic and traffic cost, and PLA presence. domain_rank answers for ONE database, domain_ranks answers for every database at once (one row per country), and domain_rank_history returns the month-by-month series for trend charts. SPENDS API UNITS, per line returned, so Row Limit is a spending control and not a convenience: domain_rank costs 10 API units per line, or 50 per line when Historical Month is set; domain_ranks costs 10 API units per line, or 50 per line when Historical Month is set; domain_rank_history costs 10 API units per line. Traffic and search-volume figures here are Semrush's MODELLED estimates derived from its own SERP crawl and clickstream panel, not measured visits to the site. Treat them as directional, and never present them beside real analytics numbers as though they were the same kind of quantity. Positions, visibility and authority scores are POINT-IN-TIME readings, not counts: they cannot be added across periods, and an unweighted average of them is also wrong. That is why these nodes declare no aggregatable measures. Output shape: Semrush answers this endpoint with semicolon-delimited CSV, not JSON, so this connector parses it. data.columns is the header row, data.rows is one object per line keyed by those column codes, and data.row_count says how many came back. Values are the vendor's own text, uncast, because one report can mix decimals, packed date stamps and comma-joined trend series in a single table. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Domain Paid Search
What a domain buys on Google Ads: domain_adwords lists the paid keywords with position, CPC and ad block, domain_adwords_unique lists the distinct ad copies with their titles and descriptions, and domain_adwords_historical returns the month-by-month ad history. Useful for competitor ad-copy research and for sizing someone's paid spend. SPENDS API UNITS, per line returned, so Row Limit is a spending control and not a convenience: domain_adwords costs 20 API units per line, or 100 per line when Historical Month is set; domain_adwords_unique costs 40 API units per line; domain_adwords_historical costs 100 API units per line. Traffic and search-volume figures here are Semrush's MODELLED estimates derived from its own SERP crawl and clickstream panel, not measured visits to the site. Treat them as directional, and never present them beside real analytics numbers as though they were the same kind of quantity. Output shape: Semrush answers this endpoint with semicolon-delimited CSV, not JSON, so this connector parses it. data.columns is the header row, data.rows is one object per line keyed by those column codes, and data.row_count says how many came back. Values are the vendor's own text, uncast, because one report can mix decimals, packed date stamps and comma-joined trend series in a single table. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Domain Shopping Ads
Google Shopping / Product Listing Ads for a domain: the PLA keywords it appears on with product titles and prices (domain_shopping), its distinct product ad copies (domain_shopping_unique), and the retailers competing for the same PLA placements (domain_shopping_shopping). SPENDS API UNITS, per line returned, so Row Limit is a spending control and not a convenience: domain_shopping costs 30 API units per line, or 150 per line when Historical Month is set; domain_shopping_unique costs 60 API units per line, or 300 per line when Historical Month is set; domain_shopping_shopping costs 60 API units per line. Traffic and search-volume figures here are Semrush's MODELLED estimates derived from its own SERP crawl and clickstream panel, not measured visits to the site. Treat them as directional, and never present them beside real analytics numbers as though they were the same kind of quantity. Output shape: Semrush answers this endpoint with semicolon-delimited CSV, not JSON, so this connector parses it. data.columns is the header row, data.rows is one object per line keyed by those column codes, and data.row_count says how many came back. Values are the vendor's own text, uncast, because one report can mix decimals, packed date stamps and comma-joined trend series in a single table. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Get Project
One Semrush project by id, with its name, domain, owner, permissions and enabled tools. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Indexed Pages
The pages of a site that attract links, ranked by how many: source URL and title, backlink and referring-domain counts, the page's response code, and its internal and external link counts. This is how you find which content actually earns links. SPENDS API UNITS: 40 units per LINE. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Keyword Ideas
Keyword expansion around a seed phrase: semantically related keywords with a relevance score (phrase_related), every keyword containing the phrase (phrase_fullsearch), and the question-form variants that make good article briefs (phrase_questions). Filter server-side with the Filter field to keep the bill down, because you pay per line returned. SPENDS API UNITS, per line returned, so Row Limit is a spending control and not a convenience: phrase_related costs 40 API units per line; phrase_fullsearch costs 20 API units per line; phrase_questions costs 40 API units per line. Traffic and search-volume figures here are Semrush's MODELLED estimates derived from its own SERP crawl and clickstream panel, not measured visits to the site. Treat them as directional, and never present them beside real analytics numbers as though they were the same kind of quantity. Output shape: Semrush answers this endpoint with semicolon-delimited CSV, not JSON, so this connector parses it. data.columns is the header row, data.rows is one object per line keyed by those column codes, and data.row_count says how many came back. Values are the vendor's own text, uncast, because one report can mix decimals, packed date stamps and comma-joined trend series in a single table. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Keyword Metrics (v4)
Search volume, CPC, competitive density, keyword difficulty, result count, search intents, SERP features and the 12-month trend for one keyword, from Semrush's newer JSON keyword API. This is the vendor's recommended replacement for the phrase_this report; the CSV keyword nodes remain because v4 replaces none of the others. SPENDS API UNITS: 20 units per REQUEST, flat. Volume is a modelled estimate and difficulty is a point-in-time 0-100 score; neither is a count and neither may be aggregated. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Keyword Overview
The core metrics for a keyword: search volume, CPC, competitive density and result count. phrase_this answers for one database, phrase_all answers for every database at once, phrase_these takes a batch of up to 100 keywords in one call, and phrase_kdi returns the Keyword Difficulty Index. phrase_these is the cheap way to score a list: one call, 10 units per line, instead of one call per keyword. SPENDS API UNITS, per line returned, so Row Limit is a spending control and not a convenience: phrase_this costs 10 API units per line; phrase_all costs 10 API units per line; phrase_these costs 10 API units per line; phrase_kdi costs 50 API units per line. Traffic and search-volume figures here are Semrush's MODELLED estimates derived from its own SERP crawl and clickstream panel, not measured visits to the site. Treat them as directional, and never present them beside real analytics numbers as though they were the same kind of quantity. Keyword Difficulty is a point-in-time 0-100 score, so it must never be summed or averaged across pulls. Output shape: Semrush answers this endpoint with semicolon-delimited CSV, not JSON, so this connector parses it. data.columns is the header row, data.rows is one object per line keyed by those column codes, and data.row_count says how many came back. Values are the vendor's own text, uncast, because one report can mix decimals, packed date stamps and comma-joined trend series in a single table. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Keyword SERP and Advertisers
Who occupies a keyword's SERP: the domains and URLs ranking organically (phrase_organic), the advertisers bidding on it (phrase_adwords), and the month-by-month history of who advertised on it (phrase_adwords_historical). SPENDS API UNITS, per line returned, so Row Limit is a spending control and not a convenience: phrase_organic costs 10 API units per line; phrase_adwords costs 20 API units per line; phrase_adwords_historical costs 100 API units per line. Positions, visibility and authority scores are POINT-IN-TIME readings, not counts: they cannot be added across periods, and an unweighted average of them is also wrong. That is why these nodes declare no aggregatable measures. Output shape: Semrush answers this endpoint with semicolon-delimited CSV, not JSON, so this connector parses it. data.columns is the header row, data.rows is one object per line keyed by those column codes, and data.row_count says how many came back. Values are the vendor's own text, uncast, because one report can mix decimals, packed date stamps and comma-joined trend series in a single table. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: List Projects
Every Semrush project on the account, with its id, name, domain, owner and which tools are enabled on it. This is the node the other project nodes' Project pickers read, and the only lister anywhere in the Semrush API, because everything else takes a domain, a URL or a keyword you type yourself. Does not consume report units: the vendor's Projects API documents no per-call unit price, unlike every Analytics report. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Market Leaders and Movers
Market-wide lists rather than a single target: rank returns the domains with the highest Semrush Rank in a database, and rank_difference returns the winners and losers, the domains whose organic or paid visibility moved most since the previous period. Neither takes a domain: these are the whole-market views. SPENDS API UNITS, per line returned, so Row Limit is a spending control and not a convenience: rank costs 10 API units per line, or 50 per line when Historical Month is set; rank_difference costs 20 API units per line, or 100 per line when Historical Month is set. Traffic and search-volume figures here are Semrush's MODELLED estimates derived from its own SERP crawl and clickstream panel, not measured visits to the site. Treat them as directional, and never present them beside real analytics numbers as though they were the same kind of quantity. Positions, visibility and authority scores are POINT-IN-TIME readings, not counts: they cannot be added across periods, and an unweighted average of them is also wrong. That is why these nodes declare no aggregatable measures. The difference columns (Tm, Um, Am, Bm, Cm) are already a delta between two periods, so they must not be summed with each other or across pulls. Output shape: Semrush answers this endpoint with semicolon-delimited CSV, not JSON, so this connector parses it. data.columns is the header row, data.rows is one object per line keyed by those column codes, and data.row_count says how many came back. Values are the vendor's own text, uncast, because one report can mix decimals, packed date stamps and comma-joined trend series in a single table. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Referring Domains
The domains linking to a target, one row each, with how many backlinks they send, the domain's Authority Score, its IP and country, when it was first and last seen, and whether it is newly gained or recently lost. Usually more useful than the raw link list for outreach and for competitor gap analysis. SPENDS API UNITS: 40 units per LINE. Rows come back ordered and cut off at Row Limit, so a result is a TOP N rather than the whole profile. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Referring IPs
The IP addresses linking to a target, with backlink and domain counts and country. The report you use to spot a link network: many referring domains sharing one IP is a footprint, not a coincidence. SPENDS API UNITS: 20 units per LINE, the cheapest of the per-line backlinks reports. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Run Any Report
Escape hatch for the classic Analytics API: run ANY documented report type against api.semrush.com with whichever target and display parameters it needs. The full vocabulary this node accepts, all 39 of them: domain_adwords, domain_adwords_adwords, domain_adwords_historical, domain_adwords_unique, domain_domains, domain_organic, domain_organic_organic, domain_organic_pages, domain_organic_subdomains, domain_rank, domain_rank_history, domain_ranks, domain_shopping, domain_shopping_shopping, domain_shopping_unique, phrase_adwords, phrase_adwords_historical, phrase_all, phrase_fullsearch, phrase_kdi, phrase_organic, phrase_questions, phrase_related, phrase_these, phrase_this, rank, rank_difference, subdomain_adwords, subdomain_adwords_unique, subdomain_organic, subdomain_organic_unique, subdomain_rank, subdomain_rank_history, subdomain_ranks, url_adwords, url_organic, url_rank, url_rank_history, url_ranks. Prefer a typed node where one exists, because it validates the report type, names the columns and states the unit cost. Reach for this one for a report we grouped awkwardly or a type Semrush ships after this connector. Costs whatever that report costs, per line. Output shape: Semrush answers this endpoint with semicolon-delimited CSV, not JSON, so this connector parses it. data.columns is the header row, data.rows is one object per line keyed by those column codes, and data.row_count says how many came back. Values are the vendor's own text, uncast, because one report can mix decimals, packed date stamps and comma-joined trend series in a single table. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Subdomain Report
The domain reports again, scoped to a single subdomain such as blog.example.com or docs.example.com: overview for one or all databases, the history series, the organic and paid keyword lists, the organic pages roll-up, and the distinct ad copies. Useful when a subdomain is really its own property. SPENDS API UNITS, per line returned, so Row Limit is a spending control and not a convenience: subdomain_rank costs 10 API units per line; subdomain_ranks costs 10 API units per line, or 50 per line when Historical Month is set; subdomain_rank_history costs 10 API units per line; subdomain_organic costs 10 API units per line, or 50 per line when Historical Month is set; subdomain_adwords costs 20 API units per line, or 100 per line when Historical Month is set; subdomain_organic_unique costs 10 API units per line, or 50 per line when Historical Month is set; subdomain_adwords_unique costs 40 API units per line. Traffic and search-volume figures here are Semrush's MODELLED estimates derived from its own SERP crawl and clickstream panel, not measured visits to the site. Treat them as directional, and never present them beside real analytics numbers as though they were the same kind of quantity. Positions, visibility and authority scores are POINT-IN-TIME readings, not counts: they cannot be added across periods, and an unweighted average of them is also wrong. That is why these nodes declare no aggregatable measures. Output shape: Semrush answers this endpoint with semicolon-delimited CSV, not JSON, so this connector parses it. data.columns is the header row, data.rows is one object per line keyed by those column codes, and data.row_count says how many came back. Values are the vendor's own text, uncast, because one report can mix decimals, packed date stamps and comma-joined trend series in a single table. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: URL Report
The same analysis as the domain reports, narrowed to ONE page. url_rank and url_ranks give the page's estimated traffic and keyword counts (one database, or all of them), url_rank_history gives the month-by-month series, and url_organic / url_adwords list the keywords that page ranks or advertises for. This is the node to point at a competitor's single winning article. SPENDS API UNITS, per line returned, so Row Limit is a spending control and not a convenience: url_rank costs 10 API units per line, or 50 per line when Historical Month is set; url_ranks costs 10 API units per line, or 50 per line when Historical Month is set; url_rank_history costs 10 API units per line; url_organic costs 10 API units per line, or 50 per line when Historical Month is set; url_adwords costs 20 API units per line, or 100 per line when Historical Month is set. Traffic and search-volume figures here are Semrush's MODELLED estimates derived from its own SERP crawl and clickstream panel, not measured visits to the site. Treat them as directional, and never present them beside real analytics numbers as though they were the same kind of quantity. Positions, visibility and authority scores are POINT-IN-TIME readings, not counts: they cannot be added across periods, and an unweighted average of them is also wrong. That is why these nodes declare no aggregatable measures. Output shape: Semrush answers this endpoint with semicolon-delimited CSV, not JSON, so this connector parses it. data.columns is the header row, data.rows is one object per line keyed by those column codes, and data.row_count says how many came back. Values are the vendor's own text, uncast, because one report can mix decimals, packed date stamps and comma-joined trend series in a single table. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Action nodes 3
Semrush: Create Project
Create a Semrush project for a domain. A project is the container Position Tracking, Site Audit and the other campaign tools hang off, so this is the node that bootstraps a new client. Reversible: Semrush: Delete Project removes it. Semrush enforces a per-plan project limit and answers ERROR 521 when it is reached. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Delete Project
Delete a Semrush project. DESTRUCTIVE and not undoable through the API: deleting a project takes its Position Tracking history and Site Audit history with it, and recreating a project with the same domain starts that history from zero. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Semrush: Update Project
Rename a Semrush project. Only the project name is editable; the domain is fixed at creation, so moving a project to a different site means creating a new one. This is one of Semrush's newer v4 JSON endpoints, not the classic CSV Analytics API: it answers JSON with a meta object and a data array, and the API key is sent as an Authorization header rather than in the query string. data.rows is a flat alias this connector adds for the vendor's data array, so charts and datasets have one path to bind to. Semrush reports failures as a plain-text ERROR line with an HTTP 200 as often as not, so this node inspects the body: an ERROR response comes back with success false, the vendor's message, and error_code set. ERROR 50 (NOTHING FOUND) means the key worked and the query matched nothing.
Connecting Semrush takes one credential
The key lives in your Semrush account on the API keys page while signed in. One key covers everything this connector calls. The classic Analytics API takes it as a URL query parameter, which is its only accepted form, while the v4 endpoints accept an Authorization header, and the connector uses the header wherever it can rather than putting a secret in a URL. Access is the part to check before anything else, because there is no free tier of any kind. Semrush documents the Standard API, which covers both the SEO reports and the Projects endpoints, as requiring an SEO Business plan subscription plus a separately purchased API units package. There is no sandbox, no trial API and no free key, so a working credential means somebody holds that subscription and has bought units. Semrush publishes no per-unit price on its developer documentation, so this page does not quote one: the unit costs stated per node are unit counts, and what a unit costs you is on your own subscription dashboard. Semrush's other APIs are separate purchases with their own quota models, including Trends, Listing Management and Map Rank Tracker, and none of them is what this connector calls.
Sign in to Semrush and open the API keys page in your account. Copy the active key.
In Mattering Automations, open Connectors, click Add Connector, choose Semrush, and create a credential with that key.
Click Test Connection. Mattering asks a domain report about a domain in a reserved test-only top-level domain, which cannot resolve and therefore cannot have Semrush data. The account is authenticated and the classic report host is exercised, and because Semrush bills per line returned and no lines come back, the probe spends nothing. A pass then reads the unit balance, which is also free, and reports it in the success message.
All 28 Semrush nodes appear in the canvas palette. Run Semrush: API Units Balance first, for free, and read the balance before you build anything that loops.
Common uses for the Semrush MCP
Gate every report on the unit balance
Semrush: API Units Balance is the only node here that spends nothing, and it exists to be the first step rather than a curiosity. Put it in front of an expensive branch, compare the balance against what the next pull will cost at its stated per-line price, and stop the run rather than have Semrush answer with a zero-balance error halfway through a workflow. On a per-line API this is the difference between a report that did not run and a report that ran until the money ended, which are very different things to explain on a Monday.
Competitive research where the limit is the budget
Semrush: Domain Competitors ranks the domains sharing your organic or paid keywords, and compares up to five named domains keyword by keyword. It is also the most expensive family here, reaching 80 units per line on the widest report, so the row limit is a spending decision made in advance rather than a display preference. Semrush: Domain Organic Search and Semrush: Domain Paid Search cover the footprint of a single domain at a lower per-line rate, and Semrush: Domain Overview is the cheap summary to run first. Schedule the summary, and reach for the wide reports only when the summary moves.
Backlink monitoring on the current v4 family
The nine backlinks nodes are built on Semrush's v4 JSON endpoints rather than the older CSV family, because Semrush marks the older one as deprecated and not recommended for new integrations. Semrush: Backlinks Overview is a flat 45 units per request, which makes it the right sizing call. Semrush: Authority Score Distribution is 1 unit per line, the cheapest node in the connector, and shows how the referring domains are spread across authority bands. Semrush: Backlinks itself is 45 units per LINE, so a thousand-link pull is a real number: size the profile first, filter server-side, and only then enumerate. Semrush: Referring Domains, Semrush: Backlink Anchors, Semrush: Backlink Competitors, Semrush: Referring IPs and Semrush: Indexed Pages fill in the rest of the picture.
Keyword research that lands in a brief
Semrush: Keyword Overview scores a keyword or a batch of up to 100 in one call, which is the cheap way to work through a list rather than paying per keyword for separate calls. Semrush: Keyword Ideas expands a seed into related terms, broad matches and the question-form variants that make good article briefs, and it is one of the reports where server-side filtering pays for itself directly, because you pay per line returned. Semrush: Keyword SERP and Advertisers shows who currently holds the result page and who is bidding on it. Chain the three into a content brief that lands in a Productivity task, so the writer sees the difficulty and the competition in the same card as the assignment.
Market-level movement, not just your own domain
Semrush: Market Leaders and Movers is the pair of whole-market views: the domains with the highest Semrush Rank in a database, and the winners and losers whose organic or paid visibility moved most since the previous period. Neither takes a domain, which is what makes them useful as a category watch rather than a self-report. Run them on a schedule into a Mattering database and route a new entrant into a task for someone to look at. Two cautions the nodes carry themselves: the difference columns are already a delta between two periods, so they must not be summed, and the wider of the two reports is priced at a higher rate per line.
Project inventory and campaign setup
Semrush: List Projects is the only lister anywhere in the Semrush API, because everything else takes a domain, a URL or a keyword you type yourself. It is what fills the project picker on Semrush: Get Project, Semrush: Update Project and Semrush: Delete Project, showing each project by name with its domain alongside so two projects called Client are not ambiguous at the moment you have to choose. Semrush: Create Project is the write side, which makes campaign setup something an onboarding workflow can do rather than a manual step. Semrush: Delete Project is flagged destructive and should sit behind an approval.
Example: A monthly competitor snapshot that checks its balance first
Built on the Automations canvas in a few minutes, or described in plain language to Matter and built for you.
Trigger: a monthly schedule fires.
Semrush: API Units Balance runs first and costs nothing. A condition compares the balance against the cost of the rest of this workflow at its stated per-line rates, and stops the run and opens a task if there is not enough left.
Semrush: Domain Overview reads the summary metrics for your own domain and for each competitor at the lowest per-line rate in the domain family.
Semrush: Domain Competitors runs with a small explicit row limit, because this is the most expensive family in the connector and the limit is what the run costs.
Semrush: Backlinks Overview adds each competitor's link profile at a flat 45 units per request, and Semrush: Authority Score Distribution adds the authority spread at 1 unit per line.
The parsed rows land in a Mattering database beside last month's, a task opens only when a competitor crosses a threshold you set, and the retention policy on that database is set with Semrush's one-month caching restriction in mind.
Semrush MCP questions, answered
What does a Semrush API call cost to run?
Semrush meters in API units, and most Analytics reports bill PER LINE RETURNED. That single fact is why the row limit on these nodes is described as a spending control rather than a convenience: one domain organic call for 1,000 rows at 10 units per line is 10,000 units. The rate differs per report and every node states its own. As examples from Semrush's own documentation, the basic domain and URL reports and several keyword reports are 10 units per line, the paid-search and broad-match reports are 20, the organic and paid competitor reports and two of the keyword idea reports are 40, the multi-domain comparison is 80, and pinning a historical month multiplies the rate by five on the reports that publish a historical price. Four things price differently: Semrush: Backlinks Overview is a flat 45 units per request, Semrush: Keyword Metrics (v4) is a flat 20 per request, Semrush: Authority Score Distribution is 1 unit per line, and Semrush: API Units Balance is free. Semrush does not publish a per-unit price on its developer documentation, so this page quotes unit counts and not currency: what a unit costs you is on your own subscription dashboard.
Does this work with Claude, ChatGPT, or other AI assistants?
Yes. Mattering exposes these nodes over MCP, an open protocol for connecting AI assistants to tools, so any MCP-capable assistant, including Claude and ChatGPT, can connect to your workspace and use them. Matter, the assistant built into the platform, uses the same nodes natively. The half that matters on a per-line API is that the same nodes also run unattended on a schedule or a trigger, so the free balance check can gate the spend whether or not a person is in the loop, and the key is held once in the workspace rather than in each person's client config.
Is there a free Semrush API tier?
No, and this one is not a judgement call. Semrush's own access documentation states that the Standard API, which covers the SEO reports and the Projects endpoints, requires an SEO Business plan subscription and that users must separately buy API units through the subscription dashboard, with no free tier. There is no sandbox, no trial API and no free key. A working credential therefore means somebody holds an SEO Business subscription and has purchased a units package. Semrush's other APIs are separate purchases with their own quota models, including Trends, Listing Management and Map Rank Tracker, and this connector does not call any of them.
Does testing the connection spend units?
No, deliberately. Because Semrush bills the Analytics reports per line returned, a probe that returned lines would charge the account every time somebody clicked Test Connection, which on a metered connector would be a defect rather than a detail. Mattering instead asks a domain report about a domain in a reserved test-only top-level domain, one that cannot resolve and therefore cannot have Semrush data. The key is authenticated, the classic report host and the parameter wiring are exercised, and the answer is a nothing-found response with no lines billed. A pass then reads the unit balance, which Semrush also serves free, and reports it in the success message. What a pass does not prove is that the v4 families are entitled, since backlinks, keyword metrics and projects live on a different path with a different auth header.
Semrush publishes its own MCP server. Why use this connector?
Semrush does, and for asking questions in a chat client it is a reasonable answer. Semrush documents a hosted MCP endpoint with OAuth as the default authentication approach and API key authentication available, exposing discovery tools for domain overview, organic research, keyword research, competitor research, backlinks research, audience research, traffic overview, paid search research, shopping research, position tracking, site audit and projects, plus a report schema tool and a generic report runner. It is also gated on a different set of subscriptions than the Standard API this connector uses, so which one you can reach may not be a matter of preference. This connector is a different product rather than a worse copy: 28 typed nodes that run on a schedule or a trigger with nobody present, join Semrush data to Google Search Console and everything else in the fleet inside one workflow, hold the credential once in a shared workspace store rather than in an individual's config, flag the destructive node, and carry the per-line unit price on the node so the spend is reviewable before the run.
Can I store Semrush data in a database or a dashboard?
Within limits that are worth knowing before you design the dataset. Semrush's API usage restrictions state that you may not cache information received via the API for more than one month without Semrush's express written consent. That is a retention constraint on whatever consumes these nodes, not something the connector can handle for you: if a workflow writes Semrush rows into a Mattering database, an Analytics dataset or a Productivity widget, the retention policy on that store is the thing that has to comply. The practical shape is a rolling window with a scheduled cleanup, or written consent from Semrush for anything longer. The same restrictions page documents the rate limits, which are 10 requests per second and 10 simultaneous requests per account.
Why did a node report failure when the HTTP status was 200?
Because Semrush frequently reports failures as a plain-text error line rather than as an HTTP status, and reading only the status would turn a wrong key, an API that is not enabled on the plan, or an exhausted unit balance into what looks like an empty report. Every node here inspects the body instead. The codes worth recognising are the credential ones, where the key is malformed or does not match the account, the one meaning the key is real but the plan does not include API access, the one meaning the balance is zero, and the account request cap. The nothing-found case is handled separately and deliberately: it means the key worked and the query matched nothing, so the node still reports failure, because a workflow branching on success must not read no-data as data, but it also sets a distinct flag so a consumer can branch on the difference.
What Semrush operations are supported, and what is left out?
28 nodes: 25 read and 3 act, with one flagged destructive. On the classic Analytics API there are eleven typed report nodes covering domain overview, domain organic search, domain paid search, domain competitors, domain shopping ads, keyword overview, keyword ideas, keyword SERP and advertisers, URL report, subdomain report and market leaders and movers, with every one of the 39 documented report types belonging to exactly one of them, plus Semrush: Run Any Report which accepts all 39 and forwards anything Semrush ships later. On v4 there are nine backlinks nodes, Semrush: Keyword Metrics (v4), and five project nodes covering list, get, create, update and delete. Semrush: API Units Balance reads the balance for free. Deliberately not built: the older CSV backlinks family, which Semrush marks as deprecated and not recommended for new integrations and which the v4 nodes replace; the Trends, Listing Management and Map Rank Tracker APIs, which are separate subscriptions with their own quota models; and the Position Tracking and Site Audit report endpoints that hang off projects, which are additive to add later.
Put Semrush to work inside your workspace
Connect once, then use every node in visual automations and AI agent workflows across the Mattering platform.
All product names, logos, and brands are property of their respective owners and are used for identification purposes only. Use of these names, logos, and brands does not imply endorsement or affiliation. Semrush is a product of its respective owner; Mattering is an independent platform.