Skip to main content
Eli Gundry

⏱️ 4 Minutes

📝 I built an MCP server to update my HTML resume!

As I get ready to maybe do another job search, it is time, yet again, to update my resume. I don’t think I have written much about my resume, but it’s something I’ve thought a lot about, what with working at a resume writing startup for 4 years.

Put simply, it’s a static HTML resume that has a print stylesheet that is used to print it to PDF. I have it as static HTML so I can give the elements semantic meaning through schema.org’s Person type, which can represent education and employment through it’s alumniOf property. You can check out how it parses here. The resume started out as a handwritten HTML file, but is now backed by multiple Astro collections in order to make laying out this semantic data easier (and I can write Markdown, which is more fun).

In order to tailor this resume in previous job searches I’ve either:

  1. Edited the page locally to match the job
  2. Just not tailored it as I don’t like to lie or exaggerate

But those were less competitive markets and agents/LLMs should allow me to continue to be stubborn with my semantic HTML and get the job of my dreams!

Enter WebMCP

WebMCP is a web standard that Google is pushing that allows computer use agents to be able to make well defined function calls on a webpage. The main use cases they are putting forward are booking trips and making purchases, which is (sadly) on brand for what Google Ads needs, but let’s not get too depressed by that lack of imagination.

I prompted Claude to update my resume page to:

  1. Convert it’s internal representation of my resume to JSON Resume
  2. Register WebMCP tools to:
    • Update that state in such a way that it doesn’t exaggerate my experience by editing, hiding, and reordering items. Each edit has diff tracking that is stored in the hash of the URL so it can be copied around.
    • Determine if the content exceeds two pages so it can make further edits to be concise
  3. Show the tool calls made such that they can be individually reverted
  4. Make each bullet point contenteditable such that I can drop in and make small edits. These diffs are tracked in the same way the WebMCP calls are made.

With this, I can give ChatGPT/Codex, as they support WebMCP, this prompt:

Go to https://eligundry.com/resume/ and use WebMCP to update my resume to highlight my leadership and hide the section on my love-hate relationships. Hide the 3rd bullet point for Chord and update Code Climate’s description to be something that me and an early stage founder can talk about.

Below is a video of ChatGPT doing it’s magic and here’s the link to the edited resume.

I’m Extremely Excited About This!!!

I have long maintained that the Google Bot is your user #1 for accessibility. It does not have human vision and mainly relies on markup and code to understand things. A11y benefits everything, from your users that need assistance and Google to understand things.

Now, everyone has a Googlebot with all the AI chatbots essentially becoming coding agents with computer use (and in the case of ChatGPT, able to sync your Chrome cookies to it’s browser). There is absolutely no reason to not register WebMCP tools to assist the agents.

In fact, adding WebMCP is as simple as adding aria tags because it has a declarative API that allows you to annotate a <form /> so an agent can more easily fill it out. From their docs:

      <form
  toolname="supportRequestTool"
  tooldescription="Submit a request for support."
  action="/submit"
>
  <label for="firstName">First Name</label>
  <input type="text" name="firstName" />

  <label for="lastName">Last Name</label>
  <input type="text" name="lastName" />

  <select
    name="select"
    required
    toolparamdescription="Determines what team this request is routed to."
  >
    <option value="Customer happiness team">Return my purchase.</option>
    <option value="Distribution team">Check where my package is.</option>
    <option value="Website support team">Get help on the website.</option>
  </select>

  <button type="submit">Submit</button>
</form>

I suspect that Google’s Gemini integration in the browser will support WebMCP once it becomes a standard.

At work, we already have an MCP server and we are basically wiring it up to also work in the user’s browser with additional tool parameters to optionally navigate the user to a page with the data. Because the agent has computer use, they can self correct the view if things look off, which is helpful for some of the more complicated offerings we are building!

Go on and give it a try!!!