How to connect Claude or ChatGPT to your job board with MCP

If you already run your job search through Claude or ChatGPT, you have probably copied text back and forth: pasting a job description in, pasting the assistant's notes into a spreadsheet or a tracker afterwards. You can connect ChatGPT to your job search directly instead, so the assistant reads and updates your board on its own. The bridge is MCP, and setting it up takes about two minutes.

This guide covers what MCP is, how to connect Claude, ChatGPT, and Claude Code to jobwyre, and what to do with the connection once it is live.

What MCP actually is

MCP, the Model Context Protocol, is an open standard for letting an AI assistant talk to an external tool. Instead of you pasting data in and out, the assistant calls the tool directly: it can ask your job board a question, get a structured answer back, and act on it.

jobwyre runs an MCP server at https://jobwyre.app/api/mcp. Any MCP-compatible client can connect to it. The assistant never sees your login, and jobwyre never sees your chat history. All that crosses the line is the specific tool calls the assistant makes, and you decide up front whether it can only read your board or also change it.

Nothing about this is jobwyre-specific magic. MCP is the same protocol a growing number of tools expose, and the same clients (Claude, ChatGPT, Cursor, and others) speak it.

What you need first

A jobwyre account. Sign in with Google, it takes a few seconds.
The assistant you want to connect. Read-only access works on every plan. Letting the assistant add or update jobs needs the Pro or Max plan.

That is the whole list. There is no key to generate for the recommended path, and nothing to install.

Connect Claude

Claude supports custom connectors directly, which is the cleanest way in.

Open Claude's settings and find the Connectors section.
Choose Add custom connector.
Name it jobwyre and paste the server URL: https://jobwyre.app/api/mcp
Click Connect. A jobwyre sign-in page opens.
Sign in, then pick what the connector can do: read only, or read and write. Approve.

The jobwyre tools now show up in Claude. You can start asking about your search straight away.

The exact menu labels shift between Claude's desktop app, the web app, and Claude in other surfaces, but the three things you are looking for are always the same: a connectors or integrations screen, an "add custom connector" action, and a field for the server URL.

Connect ChatGPT

ChatGPT reaches custom MCP servers through developer mode.

In ChatGPT's settings, open Connectors, then Advanced, and turn on Developer mode.
Back in Connectors, choose Add custom connector (sometimes shown as "add an MCP server").
Name it jobwyre and enter the URL: https://jobwyre.app/api/mcp
Connect, sign in to jobwyre, and approve the access level you want.

Once it is connected, you enable the jobwyre tools for a conversation the same way you would enable any other connector, and then talk to it normally.

Connect Claude Code and other config-file clients

Some clients do not have a connector screen and expect a small config block instead. Claude Code is the common example.

For these, jobwyre gives you a short JSON snippet that points at the MCP server and authenticates with the same API key you use for the REST API. You will find your key and the exact snippet to paste in the dashboard, under API, and the full walkthrough is on the Connect your AI page. Paste the block into the client's MCP config, restart it, and the tools appear.

If your config-file client does support OAuth-based remote servers, you can point it at https://jobwyre.app/api/mcp and go through the same browser sign-in as Claude and ChatGPT, with no key to store.

What to ask once it is connected

The point of connecting an assistant is that you stop doing the bookkeeping by hand. A few things that work well:

Catching up. "What should I follow up on this week?" "Which applications have gone quiet?" "What interviews do I have coming up?" The assistant reads your board and answers from what is actually there.

Logging applications. "I applied to the product designer role at Acme today, here is the link." It creates the card with the company, title, link, and date, so you are not switching windows every time you hit submit on an application.

Keeping things current. "Move the Figma interview to Thursday at 2pm." "Mark the Globex role as rejected, they went with someone internal." "Add a note to the Initech card that the recruiter is back from leave next Monday." These are the edits that never get made when they mean opening an app.

Checking limits. "How many adds do I have left today?" if you are on a plan with a cap.

jobwyre's assistant is prompted to confirm anything you did not explicitly give it rather than inventing a value, so it will ask before guessing at a date or an outcome.

Read-only versus read and write

When you connect, you choose the access level, and you can connect read-only without a paid plan.

Read-only is genuinely useful on its own. The assistant can answer any question about your search, summarise where things stand, and help you plan your week, without being able to touch the board. If you mostly want a smarter view of your own data, this is enough.

Read and write is what turns the assistant into something that maintains the board for you. It needs Pro or Max, because those write actions count against the same daily limits as the REST API.

Revoking access

Every connection you make shows up in the Connected apps list on the Connect your AI page. Each one has a one-click Disconnect that cuts access immediately. If you used the config-file path with an API key, regenerating the key in the dashboard invalidates every old config at once.

You are never locked in. Connect an assistant, try it for a week, and disconnect it if it is not pulling its weight.

Where this fits

Connecting an assistant does not replace the tracker, it sits on top of it. The board, the interview calendar, and the reminders all work on their own. MCP just means you can drive them in plain language from the tool you already have open.

If you are still deciding whether a dedicated tracker is worth it over a spreadsheet, that is a fair question with a real answer: spreadsheet versus app.