PivotHop MCP server

Measured career adjacency for AI assistants. Which careers a person's skills already reach, the gap, the salary, the licence gates — computed nightly from ~260,000 live job postings, not guessed from prose.

Ask an assistant "what can an architect become?" and today the answer is assembled from whatever it read on the web. This makes it come from measurement.

Install

Easiest — no install (Claude web/desktop/mobile, ChatGPT)

Add this as a custom connector:

https://www.pivothop.com/api/mcp

Claude → Settings → Connectors → Add custom connector. ChatGPT accepts remote HTTPS endpoints only, so this is the only option there.

Local package (Claude Desktop, Claude Code, Cursor, VS Code)

{
  "mcpServers": {
    "pivothop": {
      "command": "npx",
      "args": ["-y", "pivothop-mcp"]
    }
  }
}

Anything else that speaks MCP over stdio:

npx -y pivothop-mcp

No API key. No account. Reads public data from pivothop.com.

Tools

| Tool | Answers | |---|---| | career_routes | "What can a UX designer become?" — destinations with readiness, skills you have and lack, salary, transition time, licence gates | | skill_gap | "How far is architect from interior designer?" — the measured gap between two specific roles | | who_can_reach | "Who else could do this job?" — the employer question: which occupations already cover a role you're hiring for | | salary | Posted band (p25 / median / p75) per occupation, optionally per country | | list_occupations | All 180 tracked occupations, and which have measured routes |

What "readiness" means

The share of a destination's typical posting requirements already covered by the origin occupation's skills, measured from live postings.

It is not a probability of being hired, and not a share of candidates. A 66% architect→interior-designer readiness means two thirds of what interior design postings ask for is already in the architect skill profile.

The honesty rule

Some occupations have no measured routes — they sit below the 50-posting floor needed for a defensible number, and the count moves as the corpus and its cleaning improve. Those return an explicit insufficient_data answer saying so. list_occupations reports the live split rather than a number baked in here.

The server will not guess. A tool that invents a confident answer gets caught once and distrusted forever; one that says "I don't have enough data on brewmasters" stays useful. Licence gates work the same way: a required credential is reported as a hard gate, because no amount of skill overlap shortens a degree.

Data

Live from pivothop.com, cached 30 minutes in-process. The corpus is re-scraped and re-scored nightly, so answers track the current market rather than a snapshot frozen at publish time.

Sources: ~15 job boards and public APIs, plus US BLS OEWS wage data and observed occupational mobility flows.

Links

• Site — https://www.pivothop.com
• The Adjacency Index (citable numbers) — https://www.pivothop.com/adjacency-index
• Method and glossary — https://www.pivothop.com/glossary

MIT.