Docs7 Exposes Read-Only Document Search via WebMCP and Separately Tracks Agent and Human Queries

Context7 has added agent-facing documentation services and analytics to Docs7. In browsers that support WebMCP, documentation pages expose a read-only `search_docs` tool, allowing agents to retrieve content through structured queries instead of relying solely on the DOM, visual recognition, or traditional web crawlers. In [AI Visibility](https://context7.com/docs/docs7/ai-visibility), administrators can select a 7-, 30-, or 90-day period to view request volumes from products including OpenAI, Claude, Perplexity, Cursor, OpenCode, Meta, and Mistral, along with popular paths, source countries, and timing. Queries submitted through WebMCP are tracked separately from keywords entered by humans in the site’s search box.

Technically, WebMCP allows web pages to register tools through `document.modelContext`, with each tool specifying a name, natural-language description, input schema, and JavaScript execution function. When an agent observes a page, the browser can provide these tools to the agent alongside information such as the rendered view and accessibility tree. Notably, the [September 10 version of the specification](https://webmachinelearning.github.io/webmcp/) does not require browsers to use the MCP transport format; implementations may translate tools into MCP, proprietary function calling, or other interfaces. Docs7’s change is therefore closer to turning a documentation website into a browser-native tool provider than to deploying a separate backend MCP server.

For SDK and platform teams, the value lies in seeing which APIs agents actually ask about, whether they repeatedly retrieve the same page, and whether gaps in versions or examples are causing unusual queries. This extends documentation maintenance beyond human SEO to Agent Experience. However, the company explicitly cautions that one AI request does not represent one user, one conversation, one citation, or one answer. High traffic also cannot prove that an agent ultimately produced the correct answer, and source classifications do not identify exact model versions.

WebMCP is also currently a Draft Community Group Report from the W3C Web Machine Learning Community Group, not a W3C standard. Its security section remains non-normative and already identifies prompt-injection risks in tool descriptions and return values. Docs7’s initial use of read-only search reduces the potential for direct damage, but browser support, the accuracy of source identification, query privacy, and isolation when agents carry authenticated sessions across sites are still issues that should be monitored before production deployment.