--- title: "Making a person agent-ready with a personal MCP" description: "Why I built a public, read-only MCP around my work, projects, and availability." canonical: https://alexpolonsky.com/posts/personal-mcp/ published: 2026-09-01 --- # Making a person agent-ready with a personal MCP Why I built a public, read-only MCP around my work, projects, and availability. [Open and connect the personal MCP](https://alexpolonsky.com/mcp) I built a personal MCP for myself. It lives at [alexpolonsky.com/mcp](https://alexpolonsky.com/mcp) and gives AI assistants a small, public set of information about my work, projects, experience, and availability. The information is not pulled from my private notes, and the MCP does not pretend to be me. It can only return information I have chosen to publish. ## What it can answer MCP is a standard way for assistants to use outside information and tools. Mine has two read-only tools. `get_profile` returns a selected professional profile: the kind of work I do, some of my experience, projects, and directions I am interested in exploring. `get_availability` answers a separate question: whether I am currently open to relevant opportunities, what kinds, and how to contact me. It also includes the date when that information was last updated. Keeping these separate was intentional. My experience does not change every week. Availability can. The result is still fairly modest. It is mostly a resume you can query. Instead of reading several pages and piecing together an answer, an assistant can ask for the relevant part directly. ## Who would use a personal MCP? Probably not many people will open mine and connect it just to ask questions about me. More likely, they would use it through an assistant they already use, or an agent would use it for them. A recruiter could ask whether my experience seems relevant to a role. Someone considering a collaboration could ask for the projects or parts of my background that relate to what they are building. The MCP can return the source material, but the person asking still decides what to make of it. I suspect more people will find the idea interesting than will actually use mine, at least for now. I built it partly because I wanted to see what the idea would look like as a real public service with clear limits. ## A website that assistants can read Before building the MCP, I had already done some work on the website itself. Its pages are static HTML with stable URLs, useful headings, canonical links, and structured data. There is a sitemap, a clear crawler policy in `robots.txt`, and real 404 responses for missing pages. The main pages also have Markdown versions. If an assistant asks for `text/markdown`, the site returns the cleaner version and points back to the canonical page. `llms.txt` provides a short map of the site, and an Agent Skill tells assistants which public source to use and what not to infer. I checked the work with Cloudflare’s agent-readiness scanner and an Ora audit through Is Agentic. I used the reports as checklists, not as scores to chase. I kept the parts that made sense for a public personal site and left out capabilities the site does not offer. The MCP adds something different. The rest helps an assistant navigate the website once it reaches it. The MCP gives it a defined way to ask for information. That still leaves a harder question: how does an assistant know to look for me in the first place? A sitemap can describe the site, but it cannot put it in front of the right agent. That depends on search, directories, recommendations, and whatever discovery systems grow around agents. Publishing the MCP made that gap easier to see. ## The boundary The MCP is public, anonymous, and read-only. It has no access to my second brain, private files, messages, or accounts. It cannot edit anything, contact anyone, score a candidate, or decide whether I am a fit for something. It also is not an agent that speaks on my behalf. It returns a maintained set of facts with links and a date for anything time-sensitive. An assistant can use those facts in an answer, just as it might use a public page, but it should not invent the rest. This makes the MCP less capable than it could be. I prefer that to making a personal service whose boundaries are hard to explain. ## What might come later The current version focuses on professional information. A later version could help someone find the project most relevant to what they are working on. It could also cover things that don’t belong on a resume: albums, stand-up specials, books, restaurants, or coffee shops I would recommend, along with why. That would still use information I had deliberately published. It would not need access to all my notes or try to create a copy of me. It would be closer to a small, maintained window into my experience and, for better or worse, some of my taste. And yes, building an MCP about yourself is also a very 2026 way to show how into MCPs you are. I know. I built it because I had been thinking about personal agent readiness for a while and wanted to see what it looks like in practice. The MCP doesn’t answer every question. It showed me which parts still need work: what should be public, how an agent finds it, and where the boundary should be. [Open and connect the personal MCP](https://alexpolonsky.com/mcp) Also shared on [LinkedIn](https://www.linkedin.com/posts/alexpolonsky_making-a-person-agent-ready-with-a-personal-activity-7500513747988529152--zJF).