Let an AI agent build your resume with WebMCP

What WebMCP is

WebMCP is a proposed web standard from the W3C Web Machine Learning community group, with an early implementation in Chrome. A page registers named tools with typed inputs through document.modelContext. Agents in the browser discover those tools and call them like function calls, instead of guessing where to click.

For a resume builder, the difference is reliability. The agent does not need to find the Add experience button or wait for a form to render. It calls add-section-entry with a company, title, dates, and bullet points, and the app validates the input and returns a clear error if something is wrong.

What an autonomous agent session looks like

1. The agent reads the workspace

   get-workspace returns the saved resumes and their completion. The agent can start from an existing resume with get-resume or create a new one.

2. It creates a version for the job

   create-resume opens a fresh draft named for the role, such as “Backend Engineer, Acme”. Keeping one version per application is the normal way to use the app.

3. It writes the content

   update-personal-info, update-summary, and add-section-entry fill in contact details, a summary tuned to the job description, work history, projects, and skills. Each call renders immediately in the editor.

4. It styles the page

   recommend-templates ranks the 25 templates for the target job and explains why; set-appearance applies the choice with its designed fonts and accent color, or sets A4 or Letter paper. reorder-sections, rename-section, and set-section-visibility shape the layout for this application.

5. You review and save the PDF

   export-pdf opens the preview and the browser print dialog. You check the result and choose Save as PDF. On the Android and iOS apps the PDF downloads directly.

The full tool catalog

Every control in the editor has a matching tool. Resume tools accept an optional resumeId and otherwise act on the resume that is open. Tools use the same actions as the buttons in the interface, so agent edits autosave, appear in undo history, and look exactly like edits you make yourself. While an agent works, the editor shows what it is doing and follows it to the section it is editing.

Field reference for resume sections

add-section-entry and update-section-entry take a section name and an entry object. Dates are display strings such as Aug 2021, 2020, or Present. Descriptions accept plain text or simple HTML (paragraphs, lists, bold, italic, and links); anything else is removed.

set-appearance accepts the templates tenali, tenali-classic, oslo, vienna, kyoto, geneva, austin, zurich, sydney, berlin, boston, london, toronto, helsinki, dublin, lisbon, milan, singapore, chicago, amsterdam, seattle, denver, madrid, copenhagen, paris, the page sizes A4 and Letter, and a hex accent color such as #004aad. Call list-templates for each template's best-fit roles, ATS rating, layout, and page guidance.

When an input is invalid, a tool returns a result whose text starts with Error: and lists the accepted values, so the agent can correct the call and retry. The tool definitions and validation live in src/features/webmcp/tools.ts on GitHub.

How to try it

In Chrome, enable chrome://flags/#enable-webmcp-testing, reload Resume Maker 9000, and use an agent or the Model Context Tool Inspector extension to list and call the tools. In browsers without native WebMCP, the app installs a compatible document.modelContext itself, so automation harnesses that can run JavaScript in the page can call the same tools.

agent-browser open https://resumes.byanr.com/
agent-browser webmcp list
agent-browser webmcp invoke create-resume --params '{"name":"Backend Engineer, Acme"}'
agent-browser webmcp invoke add-section-entry --params '{"section":"skills","entry":{"name":"Go","category":"Languages"}}'
agent-browser webmcp invoke export-pdf --params '{}'

Watch the agent work, live, in your own browser

Agents usually drive a separate headless browser, which normally means you only see the result at the end. A live session connects that browser to yours so you watch every edit as it happens, and anything you change flows back to the agent. There is no remote server involved: the two browsers talk through a small relay that runs on your own computer.

1. The agent starts the relay

   From the repository, the agent runs node scripts/live-relay.mjs --resume <resumeId>. It listens only on 127.0.0.1, creates a one-time token, prints an agent link and a user link, and opens the user link in your default browser.

2. Both browsers join

   The agent opens its link in its headless browser, launched with --disable-features=LocalNetworkAccessChecks so Chrome lets a website reach the relay. The token travels in the part of the link after #, which browsers never send to a web server, and the app removes it from the address bar.

3. Edits flow both ways

   Every change is sent as the same command the editor already uses. The relay puts commands in one order and both browsers apply them in that order, so the resume stays identical even when you and the agent type at the same moment. Each person keeps their own view, and a badge shows where the other one is.

The agent can call get-live-session to see whether you have joined and which section you are looking at, and get-resume to read your latest edits.

Common questions