MCP Hit 4.5% of HN Hiring Posts. Only 46 US Profiles Spell It Out
MCP appears in 4.5% of HN hiring posts, but 88,361 US profiles say MCP and only 46 say Model Context Protocol. Here's how to source real MCP server authors.
The HN Trends November 2026 report gave MCP the biggest jump of any term: it now appears in 4.5% of "Who is Hiring?" posts, up from 0.3% a year ago. Open the matching "Who wants to be hired?" thread and you'll see MCP in stack list after stack list, so the keyword can't tell you who has built anything. If you need to hire an MCP engineer, skip the acronym and start from the servers themselves: the registry entries, the repos they link to, and the version history.
What did the November HN report actually find about MCP?
MCP showed up in 11 of the November "Who is Hiring?" posts, against 1 at the same point last year. That took it from 0.3% to 4.5% of posts, the largest climb in the report. The trend is real, but the sample behind it is small.
Some context from the same report:
- The thread is shrinking. As of Nov 14 the November thread had 244 posts, down 37% from 387 at the same point last year. August's 224 posts was the lowest since January 2014.
- AI is everywhere. 51.6% of November posts mention AI.
- Agents beat LLMs. AI agents (counting "agentic") appear in 14.8% of posts, ahead of LLMs at 10.7%. The report calls its section figures three-month averages, so they don't line up exactly with the November-only tables.
Put together: fewer companies are posting, and the ones that are want people who can wire models into real systems. MCP is the protocol for that work. An MCP server exposes a tool or data source so an agent can call it. If you're working out how to find AI agent engineers, MCP is now one of the most concrete skills you can screen for. You just can't screen for it with the word.
Why does searching "MCP" return the wrong people?
A bare "MCP" search matches far more profiles than "Model Context Protocol" does, and most of the extra matches are unlikely to be protocol engineers. In Refolk's index of professional profiles, "MCP" alone in US headlines returns 88,361 profiles. The full phrase "MCP Model Context Protocol" returns 46.
The top titles on the bare-acronym results are CTO, CEO, and Co-Founder, at companies such as Operant AI, Okta, and Haptiq. Some of those people may well work on agent infrastructure. Most of the 88,361 are probably something else. MCP is an overloaded acronym, and Microsoft Certified Professional is the obvious collision. I haven't confirmed that this explains the gap, but the gap is large enough that any search string built on the bare acronym is mostly noise.
The full-phrase results don't fix the problem either. The 46 profiles are a mixed group: three Senior Software Engineers, a Lead AI forward deployed engineer, a Product Manager, and a Regional Sales Manager.
| Query (headline match) | Country | Profiles |
|---|---|---|
| "MCP" (bare acronym) | US | 88,361 |
| "MCP Model Context Protocol" | US | 46 |
| "MCP Model Context Protocol" | UK | 4 |
| Bare vs. full-phrase ratio | US | ~1,920x |
| US vs. UK full-phrase ratio | - | 11.5x |
Source: Refolk's index, headline keyword matches, Nov 2026. Ratios derived.
Keyword versus artifact
You can see the same split in the HN candidate thread. A typical entry lists MCP as one item in a stack: "Technologies: Python, Go, TypeScript… Docker, AWS, FastAPI, Django, MCP." Another candidate describes an MCP server that lets an agent coach and query a calisthenics training app. The first tells you the person has heard of the protocol. The second names something you can open, read, and ask about in an interview.
Where do real MCP server developers show their work?
Real MCP server developers show their work in the official MCP Registry, on GitHub, and on npm or PyPI, not in their profile headlines. The registry is a public index of published MCP servers, and it's the best place to start because most entries point straight to source code.
The numbers that make the registry a workable sourcing tool:
- Scale: the registry listed 66,812 unique servers on Nov 18, 2026, up 44% from 46,531 on Oct 19 and up from 15,494 on July 8 (mcpcensus). Other censuses count differently. One found 47,267 active servers on Oct 21, so always give the source and date.
- Links to code: 81.4% of registry servers link a source repository (July 28 snapshot, Scalix).
- Language: of servers that list a package, 64.0% publish to npm and 25.6% to PyPI. Most real authors write TypeScript or Python.
- GitHub tagging: 32,944 GitHub repos carried the
mcp-servertopic on Oct 4.
Look at the mismatch. More than 66,000 servers are registered, and only 46 US headlines spell out the protocol. The people who build servers don't advertise it on their profiles. They publish it to npm or PyPI. Sourcing engineers on GitHub is the most reliable way to reach them, and that means joining three data sources (registry entry, repo, professional profile) that were never built to be joined.
Why is "has published an MCP server" a weak filter?
Publishing one MCP server proves very little, because most servers are published once and never touched again. Shipping a server once is easy. Keeping one working over months is the real skill.
Here's what the registry censuses show:
- Most servers are abandoned. In the Sept 10 snapshot of 30,375 servers, 62% were published once and never updated (dev.to analysis).
- Most publishers ship once. In the Oct 14 census, 26,339 namespaces published 43,012 servers, and 23,859 of those namespaces published exactly one. That's 90.6%.
- A few accounts dominate. In the same September snapshot, about 50 accounts published roughly a third of all servers.
- Volume can be generated. One top publisher shipped 1,505 single-function micro-tools (devtoolhub).
So the obvious filters fail in both directions. "Has a server" matches tens of thousands of one-off projects. "Has many servers" can surface a code generator instead of a deep engineer. Before you read a high server count as seniority, open five of them and check whether they actually differ.
Doing that by hand for every registry namespace is tedious, and it's why falling back to the keyword is so tempting.
What separates an MCP maintainer from a one-time publisher?
A maintainer has a track record over time: several published versions spread across months, issues closed, and pull requests merged on the linked repo. The bar is maintenance, not shipping once.
Signals to check, roughly in order of strength:
- Version history in the registry. The registry stores each published version. Several releases over months suggest the author kept the server working rather than walking away.
- Issue triage on the linked repo. Closed issues with real replies show the person supports actual users.
- Merged PRs to the reference repo.
modelcontextprotocol/servershas about 92.1k stars and about 12k forks. Most forks are not real contributions. A merged PR there went through review by the project's maintainers. - Adopting an orphaned server. Several reference servers were archived, and others picked them up. The Slack server is now "maintained by Zencoder," and Richard Meyer maintains a Puppeteer fork. Keeping a server alive after its first owner leaves is a skill in its own right.
- Third-party quality grades. Directories such as Glama and Smithery (which listed 20,098 servers on Oct 14) give outside signals. Glama grades
alialtunar/hn-hiring-trends-mcp, an MCP server built on the HN hiring threads, "A quality, B maintenance." That's a review you didn't have to write.
The project's own MAINTAINERS.md shows what a verifiable credential looks like. Lead maintainers David Soria Parra and Den Delimarsky, and reference-server maintainers such as Ola Hungerford, Cliff Hall, and Tadas Antanavicius, are listed by name and tied to code you can inspect. Treat the file as a template for the evidence you want from candidates, not as a list of people to poach.
Shipping an MCP server once proves little. Keeping one alive for months is the actual job.
Why are OAuth and remote hosting the scarce MCP skills?
The scarce part of MCP work is secure remote hosting, and especially OAuth, because most servers still use static keys or no auth at all. If your role involves exposing internal systems to agents, that's the skill to source for.
The data:
- OAuth is rare. In March 2026, only 8.5% of 3,012 registry servers used OAuth (NimbleBrain). The rest relied on static API keys, tokens, or nothing.
- Remote is now the norm. 78% of servers first published in October include a remote URL.
Here's why that matters. A local server runs with the permissions of the person running it. A remote server talks to many clients over the network, so it needs real identity, scoped tokens, and revocation. Servers are going remote fast, and the people who have done OAuth on a remote MCP server are a small slice of a pool that's already small. Set against the 11 HN posts asking for MCP, the underlying need looks less like "knows MCP" and more like "can expose internal systems to agents without creating a security incident."
A sourcing workflow for Model Context Protocol developers
To find Model Context Protocol developers who can do the job, start from the registry, follow the repo links, filter on maintenance and auth, and only then look up the person's profile. Run it in that order and you'll get a short list of real builders instead of a long list of keyword matches.
- Pull registry entries in your domain. Search the registry, Glama, or Smithery for servers that wrap systems close to yours (payments, CRM, data warehouses).
- Follow the source link. With 81.4% of servers linking a repo, most entries lead straight to GitHub.
- Drop the one-shot publishers. Remove servers with a single version and no activity after publish. With 62% never updated, that clears out most of the list.
- Check bulk accounts by hand. For namespaces with dozens of servers, read a sample before counting them as depth.
- Prioritize auth and remote transport. Search repos for OAuth flows and remote endpoints. Given the 8.5% figure, this is where your shortlist really narrows.
- Match to a person. Connect the GitHub handle to a professional profile, a personal site, or an HN "Who wants to be hired?" post.
- Open with the artifact. Your first message should name their server and a specific design choice in it. Anyone who has kept a server alive for months will notice you read the code.
Step 6 is where most manual searches stall, because a GitHub handle often looks nothing like the name on a professional profile. That gap is what Refolk closes: ask for "Python engineers who maintain a remote MCP server with OAuth," and I return the people, not the repos.
FAQ
How do I hire an MCP engineer if so few profiles mention it?
Start from code, not profiles. Only 46 US headlines in Refolk's index say "MCP Model Context Protocol," while the official registry held 66,812 servers on Nov 18, 2026. Search the registry and GitHub for servers in your domain, filter for authors who have published several versions over months, then match those authors to their professional profiles.
Is a bare "MCP" Boolean search ever useful?
Rarely. In US headlines it returns 88,361 profiles, about 1,920 times as many as the full phrase, and the top titles are CTO, CEO, and Co-Founder. Use "Model Context Protocol," a registry link, or a repo URL in your search string instead of the acronym.
What language skills should an MCP server developer have?
Mostly TypeScript or Python. Of registry servers that list a package, 64.0% publish to npm and 25.6% to PyPI. Engineers in other languages can still be strong, but most of the ecosystem, and most of what you'll be reviewing, lives in those two.
What is the single strongest signal of a senior MCP engineer?
Production-grade auth on a remote server. Only 8.5% of 3,012 registry servers used OAuth in March 2026, while 78% of servers first published in October include a remote URL. Someone who has shipped and maintained OAuth on a remote MCP server has the security and hosting skills that the 11 HN posts asking for MCP most likely need.
The data behind this
Try it on the search you came here for
Stop building boolean strings. Just describe the person.
Type one sentence. I plan the search, read GitHub, public LinkedIn and Crunchbase records, and the open web as it is right now, and hand back a ranked list with the reason next to every name.
01Describe them
One plain sentence. Role, city, stack, stage, whatever matters to you.
02I read the web live
GitHub, public LinkedIn and Crunchbase records, the open web. Not a database that went stale last quarter.
03You read the shortlist
Ranked, with the reasoning under every name. Open a profile, ask a follow-up, narrow it down.
- Staff backend engineers in NYC who shipped Rust in production
- Series A fintechs in SF under 50 people, growing headcount this year
- Maintainers of fast-growing Rust web frameworks on GitHub
- No boolean, no filters, no seat to buy. One box.
- Read at search time, so a profile updated yesterday counts today.
- Every step visible as it runs, every name with its reason.
500 free credits on sign-up. No card, no demo call. See real searches.