Serpstat MCP Guide 2026: Setup, Tools, Credits and Prompts That Work

What Serpstat MCP actually does

Serpstat MCP connects AI assistants such as Claude, ChatGPT and Gemini to live Serpstat data, so you can ask SEO questions in plain language and get answers built from real reports ^[1]. MCP itself is an open standard from Anthropic that lets assistants talk to outside tools, and Serpstat's server is the bridge between that standard and its SEO database. In practice, the assistant decides which Serpstat reports to run, runs them and writes up the answer. Serpstat's own MCP page sums up the pitch as "From 9 Manual Steps to 1 Question" ^[1]. Instead of opening a report, filtering, exporting and combining spreadsheets, you ask, and the assistant chains the calls.

The help center adds that the server keeps context between steps, so follow-up questions build on earlier answers. It covers domains, keywords, backlinks, audits and rank tracking ^[2], and it can act as well as read. The short version, as of October 2026, is simple. It is one of the cheapest full SEO MCP servers we looked at, setup takes minutes over OAuth, and the thing to watch is credits.

Which plan you need, and what each question costs

Serpstat MCP requires the Team plan or higher, because it runs on API access. Individual has no API, so it has no MCP either ^[3]. Team costs $129 a month, or $100 a month billed annually. New users can test MCP on the 7-day trial, which the MCP page says comes with all features included, so start the trial on Team rather than Individual. Connecting costs nothing extra. Every query calls Serpstat's API and spends credits based on the method called, out of the pool your plan already includes.

That pool is large. Team has 200,000 API credits a month, Team x2 has 400,000 and Agency has 2 million ^[4]. A conversation that pulls three reports of 100 rows each spends about 300 credits, by our arithmetic. Most chats cost little. The expensive work is crawling, which the credits section below breaks down. Rank tracking is the opposite case: reading tracked positions through the API costs no credits, so weekly ranking summaries through MCP are almost free.

HTTP or local: picking the connection

Serpstat offers two versions of the server, and the choice shapes everything after it. The HTTP version authorizes once via OAuth, needs no install and works on web, desktop and mobile. With OAuth, your token is managed by your AI client, so nothing is pasted anywhere, which makes it the safer choice on shared machines. The STDIO version runs a local Node.js server, keeps your token on your machine and lets you load only some tool groups through SERPSTAT_ENABLED_CATEGORIES, which the help center says reduces context window usage.

Start with HTTP. Move to STDIO for daily heavy use, or when you need a client that only supports the local version, such as Gemini CLI or the coding assistants. On the local route, Serpstat says the token is stored in a .env file on your machine and is never sent to third parties. Either way, never paste the token into a chat window. For experiments, a team sub-account with a small credit allocation keeps mistakes cheap.

Connecting your assistant

Claude Desktop and ChatGPT

A Serpstat blog post from May 2026 gives the remote setup for Claude Desktop. Open Customize, then Connectors, and add a custom connector named Serpstat SEO Tools with the remote server address from Serpstat's MCP page. Then click Add and Connect, and authorize with your Serpstat account. Since February 13, 2026 you can connect ChatGPT the same way, through a custom connector with the same address and the same setup logic. The MCP page lists ChatGPT on web, desktop and mobile for the HTTP version.

Gemini CLI, Claude Code and Cursor

Gemini CLI support arrived in November 2025, when Serpstat said its server was no longer Claude-only. Gemini CLI uses the local STDIO version, and the help center lists Claude Desktop and Gemini CLI as its tested clients. The STDIO version also works with Claude Code and Cursor. That puts Serpstat data inside the coding assistant, so a developer can pull keyword data while writing a page.

Installing the local server

You need Node.js 18 or higher, your API token and an MCP-compatible client ^[2]. The help center says setup takes under five minutes and needs no developer experience. You install the package with npm install -g @serpstat/serpstat-mcp-server, and the config entry runs it through npx with your token in the SERPSTAT_API_TOKEN environment variable. For Claude Desktop, the config file sits at ~/Library/Application Support/Claude/claude_desktop_config.json on macOS and in %APPDATA%/Claude/ on Windows. Add the Serpstat entry, save and restart Claude Desktop.

Editing that JSON file is the only fiddly step. Back it up first, and watch the commas. To test the connection, the help center suggests asking for domain info on example.com; if SEO data comes back, setup is complete. That call spends a few credits, and it is the cheapest proof that the token, plan and client all work. Update the package now and then, because Serpstat adds methods often.

The tools, group by group

The MCP page says 63 SEO tools are available. The help article dated May 28, 2026 says all 65 tools across 9 categories are enabled by default, and the February 2026 changelog said 65-plus. We take 65 as current. Either way, the AI picks the tool. The groups are domain, keywords, backlinks, url, audit, rt for rank tracking, projects, page-audit and credits. In August 2026 Serpstat connected three new Rank Tracker methods to MCP: project summaries, position diffs and URL cannibalization.

• Domain: domain info, competitors, ranking keywords, top pages, history, unique keywords between domains and market research by category. Start every client audit here.
• Keywords: related keywords, suggestions, volume and CPC, top-100 results, keyword competitors and intent filtering. Intent filters are where MCP saves the most clicks.
• Backlinks: summaries, active and lost links, referring domains, anchors, top pages, intersections and threat links.
• URL: traffic and keywords for URL patterns, URL competitors, ranking keywords and missing keywords, which is page-level content gap work for editors updating old posts.
• Audit: start, stop and configure full audits, return results by issue category, compare reports and list pages with a given error.
• Rank tracking: list projects, check parsing status and return keyword and URL ranking history.
• Projects: create, list and delete projects, the containers for rank tracking, audits and historical data.
• Page audit: scan single URLs with JavaScript rendering, track progress and compare scans.
• Credits: check API usage and remaining quota across modules.

Credits, warnings and rate limits

The help center lists what the main actions cost. Domain keyword lookups run at 1 credit per row. Site audits cost 1 credit per page, or 10 with JavaScript rendering, and one-page audits cost 10 credits per scan ^[2]. That makes a single page audit cheap for a landing page check before launch: run one, fix, run again. Full audits are another matter. A JavaScript audit of 5,000 pages costs 50,000 credits, a quarter of Team's monthly API pool, by our arithmetic. Ask the AI for an estimate first.

The server will warn you before executing high-credit operations, and those warnings deserve a careful read. An assistant can misjudge scope, asking for every keyword of a big domain when you wanted the top 50. Set row limits in your prompt. It also helps to ask how many API credits you have left at the start of a long session, which prevents surprises halfway. Rate limits rarely bite in chat. Most Serpstat plans allow 1 request per second, while Agency allows ten. The February 2026 changelog says complex queries split into workflows of 10 to 13 steps, so at one call a second they finish in seconds.

Prompts and workflows that work

Good prompts name the tool, the domain, the database and the limit: Use Serpstat SEO Tools, Google US, top 20. Ask for a format such as an executive summary or a table. Serpstat's blog explains why naming the tool matters: it makes the assistant call Serpstat instead of answering from general knowledge. Without it, a chatbot may answer from memory, and memory is out of date. Vague prompts make the AI guess. Guesses spend credits.

Content gaps and domain comparisons

The help center's content gap workflow runs three prompts. First find your top competitors, then list keywords where a rival ranks 1 to 10 and you do not, then filter to volume above 1,000 and difficulty under 45. The result is a prioritized content-gap list with no tab-switching and no spreadsheet work. A December 2025 blog post suggests asking for organic visibility and keyword counts for three domains at once, then asking which keywords one ranks for that the others miss. The second question needs no repeats.

Keywords, backlinks and audits

The same post walks through generating commercial keywords for a seed term, pulling question keywords that start with how, what or why, and then asking which have CPC over $5. For backlinks, ask for domains that link to two competitors but not to you, then filter by Domain Rank, which runs from 0 to 100. Ask the assistant to drop spammy-looking domains and explain each pick, then check the top ten by hand. For audits, start one and ask for all high-priority errors; the MCP page's example compares critical errors with last week to show what is new and what got fixed.

Rankings and client reports

Ask how rankings changed this week and which keywords moved five or more positions, or request 30 days of ranking history. Since rank reads cost no credits, run these weekly. The MCP page's sample report prompt asks for a monthly SEO report with a visibility trend, top movers and three recommendations, ready to paste into an email. Check the numbers before you send.

Skills, stus.ai and custom GPTs

Serpstat Skills for Claude, released March 19, 2026, are sets of instructions that pull data through MCP and produce complete SEO reports. They are hosted on GitHub, import in one click and run across all Claude accounts on a team. The changelog lists five: Quick Wins, a full domain analysis with an action plan, competitor reverse engineering, a months-long content strategy for a site and region, and a competitive gap. Each is a standard agency deliverable, and each still needs a human edit. Because Skills need MCP, they need Team too.

Two other routes sit next to MCP. Serpstat's product pages advertise stus.ai, 10 AI agents with commands such as @contentgap and @nicheanalysis, as a separate front end on Serpstat data; we found no pricing for it. ChatGPT users who do not want MCP can use the Serpstat SEO Tool GPT, which spends API credits, and a free SEO Copywriting Brief Generator. Custom GPTs need a paid ChatGPT plan. MCP is the more flexible route.

Write access, context and the limits

Serpstat MCP can act, not only read. It can start audits and create or delete projects, so an assistant can set up a new client project from a sentence. That is useful. It also means a careless prompt can delete a project, so read what the AI proposes and confirm destructive steps yourself. Context has limits as well. The server holds conversation context between requests, but Serpstat's blog adds that this lasts only until the length limitation is reached. Long sessions lose early results. Start a new chat for each client.

Serpstat is clear that MCP works alongside the platform and "is not a replacement for the full platform interface" ^[2]. Visual dashboards, report exports and UI configuration stay in the app. Use MCP for questions and the app for setup. When no data comes back, the usual causes are a token without API access, an Individual plan or exhausted credits. A JSON typo stops the local server loading, and a missing database code gives US data by assumption. Check the credits tool first.

How it compares with Semrush, Ahrefs and SE Ranking

Semrush MCP connects ChatGPT, Claude, Gemini, Perplexity, Cursor, VS Code and Claude Code, through OAuth or an API key, with read-only access. Semrush lists MCP access from Starter at $199 a month, while Serpstat's starts at $129. Semrush has more clients. Serpstat costs less, and it can write. Ahrefs says its MCP is included on all paid plans from Lite, with 100 rows per request and 200,000 API units a month on Lite. Serpstat allows up to 60,000 rows per request on Team. Ahrefs wins on backlink data; Serpstat wins on rows.

SE Ranking includes MCP access with 25,000 API credits on Core at $129 a month, and it publishes SEO, marketing and content Skills for Claude. Serpstat Team includes 200,000 credits at the same monthly price. These two are close rivals, and Serpstat has more credits. Serpstat's own May 2026 blog compares five SEO MCP servers and recommends itself for budget-conscious teams. It is a vendor comparison, and some rival prices in it differ from the rivals' current pages, such as a Semrush price of $139.95. Use the rivals' own pages.

Who should use it, and a first week

Serpstat MCP suits SEO teams already on Team who live in Claude or ChatGPT, agencies producing repeat reports, and developers who want SEO data inside Claude Code or Cursor. If you ask more questions than you open reports, it will save hours each week. Skip it on Individual, where there is no API. Skip it if you need backlink data at Ahrefs depth, or prompt tracking for AI visibility, which is the job of LLM Brand Monitor. Skip it if nobody on the team checks AI output.

If it fits, a first week is enough to decide. On day one, start the Team trial and connect over HTTP. Day two, run the three-prompt content gap on your own site. Day three is an audit triage and day four rank tracking summaries. On day five install a Skill, on day six count the credits spent, and on day seven decide. We relied on Serpstat's documentation rather than hands-on use, and we did not test the Skills or stus.ai. We also could not confirm whether the tool count is 63 or 65 today. For the raw endpoints behind all of this, see our Serpstat API guide.

The bottom line

Serpstat MCP is a strong, low-cost way to put full SEO data inside an AI assistant, with write actions and generous rows. It needs Team and careful prompts. Connect over HTTP, ask for domain info on your own site, then run the three-prompt content gap. Limit rows and name the tool in every prompt, and check credits weekly. If the answers save you a report a week, keep it. For the plan math behind the Team requirement, read our Serpstat pricing guide.