Use the board from your AI assistant

frontierroles runs a free MCPserver. Connect it once and Claude, ChatGPT or your editor can search every open role on the board, read full adverts, and answer questions about the market — without you leaving the conversation.

Endpointhttps://mcp.frontierroles.com/mcp

Streamable HTTP, no authentication. No account, no API key, no rate card. Every role links straight to the employer's own application page.

Add it

Claude — web, desktop or mobile

Open Settings → Connectors
Choose Add custom connector
Paste the endpoint above and save

Leave authentication blank — the server doesn't use any. Switch the connector on for a conversation from the + menu beside the message box.

ChatGPT — web

Open Settings → Security and login and turn onDeveloper mode
Go to Settings → Plugins, press +, and create an app with the endpoint above as its server URL and authentication set toNone
In a chat, open the + menu and switch on the app you just created

Custom MCP servers need a paid plan and the web app. Deep research is a separate route that only accepts servers exposing tools named search and fetch, which this one doesn't — developer mode is the way in.

Claude Code

claude mcp add --transport http frontierroles https://mcp.frontierroles.com/mcp

Add -s user to make it available in every project rather than this one./mcp in a session lists what connected.

Cursor, Windsurf, VS Code and others

Add the server to your client's MCP config file:

{
  "mcpServers": {
    "frontierroles": {
      "url": "https://mcp.frontierroles.com/mcp"
    }
  }
}

VS Code is the odd one out: it wants servers rather thanmcpServers, and "type": "http" alongside the URL.

Nothing connecting? Open the endpoint in a browser. It answers with a short plain-text description rather than a web page, which is how you know the URL is right and reachable;/healthreports what the server is currently serving.

What it can do

Tool	What it does
`search_jobs`	Find roles by free text, skill, cluster, seniority, location, pay floor or how recently they were posted. Compact records, up to 50 at a time.
`get_job`	One role in full, including the advert text and the link to apply.
`get_company`	Every open role at one company, by slug or display name.
`board_stats`	Counts and median pay across the whole board, grouped by skill, cluster, seniority, country, city, company or work type. Takes the same filters as search.
`list_skills`	The skill and cluster names the filters accept, so a question doesn't quietly match nothing.

Every tool is read-only, and says so in its own annotations — nothing it can do changes anything, here or anywhere else, and clients that stop to confirm write actions have no reason to interrupt you.

Things to ask it

Find senior RAG engineering roles in Europe that publish a salary.
What do staff-level AI engineering roles pay in the UK compared to the US?
Which skills show up most often in roles paying over $200k?
Show me everything Anthropic has open right now.
I know Python and PyTorch but not much about agents. What's out there for me, and what would I need to learn?

The last one is the point, really. A board can filter; an assistant that can see the whole board can tell you what the filters would have hidden.

About the data

First-party only. Every posting comes from the employer's own applicant tracking system — Greenhouse, Lever, Ashby, Workday and a dozen others. No recruiter reposts, no scraped aggregators, no sponsored placement.
Rebuilt nightly. Roles drop off when they close or pass 90 days old, so the board doesn't accumulate listings you can no longer apply to.
Deduplicated. Employers often open one requisition per office for the same role. The server returns one.
Dated, not live. Every response carries the snapshot date, so an assistant can tell you how fresh its answer is instead of implying it's real-time.
Nothing stored. The server takes no account, sets no cookie, and keeps no record of what was asked or who asked it. It serves public job data that is already on this site.

Just want the data?

The MCP server reads two ordinary JSON files. You're welcome to read them too — no key, no sign-up.

/mcp-index.json — every open role without its advert text, plus the skill taxonomy and currency rates.
/mcp-jobs/<slug>.json — one role, with its full advert.

There's also an RSS feed if you'd rather just watch for new roles.