--- title: "I built seven MCP servers in evenings. It proves little." url: https://rytovdigital.com/blog/seven-mcp-servers-in-evenings published: 2026-09-19 updated: 2026-09-19 author: Oleksii Rytov --- The loudest excitement about AI comes from people who have nobody to convince. I am one of them, and this post is mostly about why that matters. Last month I built seven MCP servers for my own work. Five to ten evenings, I did not count properly. No proposal, nobody to ask. When I got something wrong I rewrote it the same night. ## What they are Four are for the consultancy. One writes and files posts on this site. One queues social posts and reads the numbers back. One keeps finance entries and renewal dates. One holds decisions and past events, and that is the one I use most, because otherwise I re-argue the same thing with myself every few weeks. The other three are for the photography side of our company: its CMS, its enquiry inbox, its settings. None of them is big. A thin layer over a database, a few HTTP calls, and a list of things the assistant is allowed to do. Most of the time went into that list. Deciding that the social server can schedule a post but that publishing it needs me took longer than writing either part. The plumbing was already there. I had moved my account to a Team plan a few weeks earlier and [reconnected all the API keys by hand](/blog/claude-max-team-plan-migration), so I knew where everything was. Starting cold, those evenings would have gone on that instead. ## Why it went fast Because almost none of the usual cost was there. Nobody approved a budget. There was no security review and no question about which data leaves the company, since it is my data and my company. I wrote no documentation at all and nothing went wrong, because I am the only person who reads any of this. At one point the finance server was returning the wrong shape. I changed the schema that evening and forgot about it. In a company that is a migration, a deploy window, and a conversation with whoever built against the old version. So "seven servers in ten evenings" is not really a number about MCP. It is a number about working alone. ## What it did cost Five to ten evenings is a guess, not a measurement. Untracked work is usually longer than it feels, so the real figure is probably at the top of that range or past it. There is no documentation and no tests. Nobody else could fix any of it, and if I left it alone for three months I would be reading my own code as a stranger. For tools with one user I am fine with that. If a client ran their week on one of these it would not be fine, and building it properly would have taken several times longer. That gap has nothing to do with the technology. ## What I stopped saying I used to look at a company taking months over something like this and think they were behind, or scared of it. Now I think they are pricing something I never had to pay. The model is the cheapest line on my list, and it is probably the cheapest line on theirs. The rest of their list is people. Who agrees. Who maintains it after the person who built it moves on. And what happens to the person whose job gets smaller, which somebody has to say out loud in a meeting. Whether they are right to move slowly I have no idea. Maybe being careful now is the expensive choice later. But I have stopped treating my own speed as evidence of anything except that I work alone. ## If you are comparing yourself to a demo Most demos, mine included, are built in these conditions. One person, one user, no review, no cost to getting it wrong the first time. That does not make the demo fake. It means the demo measures how long the code takes, and the code was probably never your slow part. Before you take somebody's number as your own, ask them who else had to agree and who maintains it afterwards. If the answer is nobody and them, the number does not transfer to you. ## The part I cannot speak to I have never run one of these inside a company as the person responsible for it. So I do not know which of those costs is actually the big one. I would guess it is the one about jobs getting smaller, but that is a guess from outside and I would rather leave it as a guess.