semrush-ai-tool gives Claude, Cursor and other assistants direct access to Semrush data through an MCP server

Semrush's data used to travel through the browser: run a report, export a CSV, paste it into a chat window, ask the assistant what to do with it. On 25 September the developer springvoiceswell published semrush-ai-tool on GitHub, an MIT-licensed package that gives any MCP-compatible AI client direct access to the Semrush API and returns SEO analysis in the same conversation. The repository has 111 stars and 1 fork in two days.

The project ships two entry points. The MCP server, semrush-ai-mcp, connects Claude Desktop, Cursor, Windsurf, Cline and any other client that speaks the Model Context Protocol, and exposes 18 Semrush tools. The CLI, semrush-ai, does the same job from a terminal for developers and agents that prefer a shell.

What the 18 tools cover

The list, taken from the project's own README, groups into six areas. Keyword research covers search volume, cost per click, competition scores, related keywords and question keywords from the People Also Ask panel, plus batch keyword-difficulty scoring for up to 100 keywords in one call. Domain analytics returns organic and paid keywords, traffic estimates, rank history and visibility for any domain in any of the country databases Semrush publishes, which the README names as 100 or more.

Competitor analysis discovers organic and paid rivals and pulls back the keyword lists behind them. Backlink analysis returns profile overviews, individual backlinks with anchor text, and referring domains sorted by authority. Traffic analytics reports visits, unique visitors, bounce rate and session duration through Semrush .Trends. SERP snapshots show which domains and URLs rank for a given keyword. A small helper tool reads the account's remaining API-unit balance before a heavy job runs.

Whose key it uses, and where the data goes

The MCP server runs on the operator's own machine. The Semrush API key stays there and is never sent to the AI provider directly, only to Semrush. Requests to Semrush are rate-limited by the package. The assistant receives Semrush's response, then reasons over it inside its own model. Nothing about the project itself phones home.

The trade to know before installing is the account cost. Semrush charges API units per call, and a broad prompt across many tools can consume units quickly. That is why one of the 18 tools is a units-balance reader, so an agent can check the account before starting a large batch. springvoiceswell publishes no benchmarks for that cost, and neither does Semrush; readers who want to size a run should sanity-check on a small keyword list first.

The audience for the item is broader than a search engineer. A founder who pays for Semrush can ask their assistant "which of my competitors' organic keywords do I not rank for at all, and which of them convert on commercial intent" and get a shortlist without a spreadsheet round trip. A marketer can ask "what backlinks do the top three ranking pages for this term share that I do not have" and act on the answer in the same session.

The wider point applies to any paid data source with an API. When the model can call the vendor directly, the workflow shortens: fewer copy-pastes, fewer version-drifted exports, less time spent transferring data between the panel that has it and the assistant that reasons over it. semrush-ai-tool is one instance of that shape, arriving through MCP. Once the pattern is common, the marketing operations job changes.

Source

• Repository: springvoiceswell/semrush-ai-tool on GitHub
• Model Context Protocol reference: modelcontextprotocol.io