What is a job search MCP server and how do I connect one?

Q: What's the server URL?

`https://findyourrolefirst.click/mcp`, with an `Authorization: Bearer YOUR_API_KEY` header.

2026-09-28

A job search MCP server is a service that exposes job listings as tools an AI agent can call, so Claude Code, Codex CLI or another MCP client can search open roles directly instead of scraping web pages. Find your role first runs one at https://findyourrolefirst.click/mcp, backed by 8,519 company career boards on Greenhouse, Lever and Ashby.

If MCP is new to you, the first section covers the idea in plain terms. If you just want to connect, skip to "How to connect a job search MCP server".

What is MCP, in plain terms?

MCP, the Model Context Protocol, is a standard way for an AI agent to use outside tools. A server describes the tools it offers, what inputs each one takes and what it returns. A client, like Claude Code or Codex CLI, reads that description and calls the tools when they fit the task.

The practical upshot: you connect a server once, and from then on you can just ask your agent for something in English. It figures out which tool to call.

What does a job search MCP server do?

It turns "find me senior backend roles in Berlin posted this week" into a structured search against a job database, and hands the agent back a clean list.

Without one, an agent looking for jobs has to use general web search. That works, sort of. It finds aggregator pages, older reposts and roles that already closed, and it spends a lot of tool calls getting there. A job search server gives it a direct line to the data.

Which tools does this server expose?

Find your role first has four tools:

search_jobs searches by title, company and location. It can filter by source (Greenhouse, Lever or Ashby), by publication date, and by the date the feed first discovered the role.
get_job refreshes one role's record, including whether it has closed.
get_changes returns new, updated, reopened and closed roles since a saved cursor, which suits scheduled checks.
get_usage shows how much of your monthly allowance is left.

All four are read only. None of them can apply to a job or touch anything on your behalf.

What does it return?

Each role comes back as a short record: title, company, location, dates, the original posting link and the apply link.

It doesn't include the full job description. Your agent reads that from the original posting with its own web or browser tool, and only for the roles it cares about. Honestly, it's a fair question why. Shipping full descriptions would make every search response huge, and most of that text gets thrown away. Keeping the records small means an agent can scan a lot of roles cheaply and spend its context on the few that matter.

Why publication time and discovery time are separate

Every record carries two times. One is when the company says it published the role. The other is when this feed first saw it.

They can differ. A board might show an older date on a role that only just went live, or a company might edit a posting and bump its date. If you want what's genuinely new, ask for "discovered since". If you want the company's own date, ask for "published since". Your agent can use either.

How to connect a job search MCP server

First, sign in with Google at findyourrolefirst.click/account, subscribe, and create an agent key. Then pick your client.

Claude Code

claude mcp add --transport http --scope user jobfeed https://findyourrolefirst.click/mcp --header 'Authorization: Bearer YOUR_API_KEY'

Type /mcp inside Claude Code to confirm jobfeed is connected.

Codex CLI

export JOBFEED_API_KEY=YOUR_API_KEY
codex mcp add jobfeed --url https://findyourrolefirst.click/mcp --bearer-token-env-var JOBFEED_API_KEY

Any other HTTP MCP client

Add a server entry with:

URL: https://findyourrolefirst.click/mcp
Header: Authorization: Bearer YOUR_API_KEY

Most clients take this as a JSON block in their MCP settings. The exact shape varies by client, so follow its docs for where the URL and headers go.

Things to check before picking any job search MCP server

Whichever job data server you end up with, it helps to ask:

Where does the data come from? Company career boards, or an aggregator that copied them?
How often is it checked? Every few hours, or once a day?
What does a repeat cost? Agents re-run searches a lot. If every repeat counts against you, that adds up.
Does it close roles? A feed that never marks closed roles leaves your agent chasing dead links.

For this one: boards come straight from Greenhouse, Lever and Ashby, each is checked every two hours, a role you've already seen that month is free to read again, empty searches are free, and closed roles are tracked.

What it costs

$9 a month for 3,000 new roles, up to 1,000 new a day. Cancel anytime, with a refund within 3 days.

How to try it

Sign in with Google at findyourrolefirst.click/account, subscribe, create an agent key, and paste the one line for your client from above. Then ask your agent for roles the way you'd ask a friend who knows the job market.

For client specific walkthroughs, see finding jobs with Claude Code and searching for jobs with Codex CLI.

Questions