# Semrush in Codex: A Real Keyword Research Call via BrowserAct MCP > Source: > Published: 2026-10-28 06:31:08+00:00 I had Semrush data in one tab and a content brief taking shape in Codex. To see whether I could stop moving reports by hand, I tested an existing BrowserAct Semrush Bot as an MCP tool. It requested a US Keyword Overview for “OpenAI Codex” and returned one keyword record plus five search-result links. The useful change was getting a defined research task—and its evidence—into the place where I was deciding what to write. What interested me was using the research where the decision was happening. If I was discussing an audience and a proposed article in Codex, I wanted relevant search data in that conversation—not another summary based on whatever the model remembered. I used `OpenAI Codex` as the research topic for a real BrowserAct MCP request. The idea becomes clearer with one recurring request: retrieve a keyword overview before deciding whether to develop a topic. In the browser, a person opens the report and collects the results. With BrowserAct, a published Bot can perform the supported browser task using the connected account and return structured data. The work can then start from the application or Agent that needs it. There are two ways to use that published Bot. An application can submit a task through BrowserAct API and retrieve the result after it finishes. A compatible Agent can call it as an MCP tool. This is what I mean by turning a Semrush subscription into API/MCP access: making supported web research callable through BrowserAct, rather than unlocking every endpoint in Semrush’s official API. Semrush also offers an official MCP connection on eligible subscriptions. I used BrowserAct for a different reason: to make a specific browser research task available as my own reusable Bot through MCP or the BrowserAct API. For me, the value is getting more use from a subscription I already have: supported research can feed applications and Agents instead of staying in a browser report. It reduces repeated copying and gives teammates a research tool they can reuse. If the only reason to upgrade a plan or buy separate interface access is to make that connection, BrowserAct is another route worth comparing on total cost before buying more access. The starting point is a [ready-built Semrush Bot template](https://www.browseract.com/template/semrush-scraper-copy-before-use?co-from=blog-Semrush-Codex-Research-DEV), not a blank project. Copy it into your personal BrowserAct account, import your Semrush cookies in Browser settings, and publish your copy to start running supported research tasks. There is no scraper to write or Bot to build from scratch. The template provides the task; your own Semrush login provides the access. In the Codex session used to prepare this article, I called the existing BrowserAct Semrush Bot through MCP to request Keyword Overview for `OpenAI Codex`, using the United States database. The question was concrete: before commissioning content around this topic, what does the returned search overview suggest we should investigate? *The actual Codex MCP call record, formatted as a readable excerpt. The request fields and returned values are unchanged.* The task finished successfully on October 27, 2026. It returned one exact-match keyword overview and a separate section containing five search-result links. Those are different kinds of output: one researched keyword, plus pages associated with its search results—not five keyword ideas. Here is a compact excerpt from the completed response. *The BrowserAct task record for this MCP call, with unrelated columns and account UI hidden for readability. The returned data is unchanged.* | Field | Value returned | |---|---| | Keyword | `OpenAI Codex` | | Database | United States | | Estimated monthly search volume | 14,800 | | Keyword difficulty | 94 | The returned page links included `github.com/openai/codex`, `openai.com/codex/`, `openai.com/index/introducing-codex/`, `platform.openai.com/codex`, and `learn.chatgpt.com/docs`. These links and the figures above are what the tool returned in this request. The volume is an estimate, not visits to a site; I have not independently checked every figure or opened all five pages. My first reaction was that the broad topic deserved investigation, but the volume did not tell me what article to write. The difficulty value also discouraged treating this as an obvious target. More importantly, the returned URLs pointed toward the product, its repository, and related learning material. That gave me a reason to question whether a general overview would help our audience. For marketing readers, I would investigate a narrower question: how could Codex use current SEO research while preparing a content brief? That is a business use case we can explain with this connection. It is a candidate angle, not a conclusion that the broad keyword’s demand belongs to that narrower topic. Search intent and demand for the specific angle still need checking. That gives a content meeting something useful to work with: the queried term, its returned overview, the types of pages to inspect, and a more focused research question. Instead of turning one large number into a publishing decision, the team can decide what evidence would justify the proposed angle. Having the records inside Codex makes that discussion more specific. An editor can ask why a term was selected, challenge the interpretation, or request a different supported query. The Agent has something concrete to work from, while the team keeps responsibility for choosing the topic. That is the business value I see in making an existing subscription callable. Consider the next brief. A marketer needs the same kind of overview for a different keyword. Instead of asking the SEO lead to retrieve and forward another report, they can request the supported task through the shared tool. Changing an input can reuse the Bot; it does not require writing a new scraper for each brief. BrowserAct supports privately sharing an exposed MCP tool with trusted teammates through the MCP Server URL and matching API Key. The owner maintains the connected browser login; teammates call the tool from their compatible Agents without receiving the Semrush web password. Account access rules and available capacity still govern what can be requested. For me, the team benefit is a clearer division of work. The marketer frames the question. The Bot retrieves the supported records. The SEO lead reviews the interpretation instead of becoming the courier for every report. Agreeing on the target, market, and scope keeps the result understandable when it moves between people. I would still open Semrush to explore freely or investigate an unexpected result. BrowserAct is useful here because the request is defined well enough to repeat. The real call demonstrated retrieval into Codex; the follow-up checks and team routine described above are ways to use that connection, not additional runs completed in this example. Once your personal copy is ready, the same Bot can become the tool your Agent calls. The [MCP connection guide](https://docs.browseract.com/integrations/mcp?co-from=blog-Semrush-Codex-Research-DEV) explains how to expose the published Bot and connect a compatible client. That setup is separate from the research itself: the useful test is whether the returned evidence helps your team make the next decision. If the task you need is on another website, [BrowserAct](https://www.browseract.com/?co-from=blog-Semrush-Codex-Research-DEV) also lets you describe the data you want in natural language and build a reusable Bot without writing code. Describe the source, the records to collect, and the fields your team needs. Once the Bot is built and tested, you can run it directly or publish it and connect it to a compatible Agent through MCP. The website’s login and access requirements still apply. If you already use Semrush and want its supported research available inside an application or Agent, [try turning one recurring task into an API/MCP tool with BrowserAct](https://www.browseract.com/template/semrush-scraper-copy-before-use?co-from=blog-Semrush-Codex-Research-DEV). This Semrush-in-Codex case shows the starting point: request records from the subscription you already use, examine what came back, and make the next research decision with evidence in front of you.