Context Over MCP, Part IV: No Working Directory At All
Part I: an MCP client has no working directory, so it never sees your AGENTS.md . Part II: publish the server so a client can find it. Part III: keep facts as data and instructions as prose, so what it finds is worth seeing. Now take the working directory away completely. WebMCP's client is a browser tab, and a tab doesn't fail to find a working directory. It doesn't have the concept. So what…
A WebMCP client operates within a web browser tab, which inherently lacks a concept of a working directory. This absence eliminates the need for an MCP process, transport, or working directory within the client. To register tools and expose context in this environment, WebMCP relies on Chrome's experimental API and a fallback polyfill package, @mcp-b/webmcp-polyfill.
Three tools are demonstrated: fill_6ws, score_faf, and emit_agents_md. fill_6ws is a simple HTML form that registers itself as a tool in the browser. score_faf, a client-side scoring kernel written in WASM, evaluates the provided YAML data and returns a score ranging from 0 to 100. emit_agents_md generates a minimal AGENTS.md file from the scored facts and presents it to the user.
The registration process involves setting the appropriate attributes on the HTML form, such as toolname, tooldescription, and toolautosubmit. If the browser fails to automatically detect the form, the script registers it manually. The scoring kernel runs entirely in the browser, eliminating the need for an external server or MCP process.
The tool also refuses to inflate its score to appear better, maintaining honesty about its limitations. The emitted AGENTS.md file is a simplified version, explicitly stating it is not the complete output produced by the CLI process.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
