Looking for a job with an AI agent: practical guide

Job searching is an unpaid part-time job: browsing the same sites every morning, retyping the same filters, sorting the offers already seen, locating those published since the day before. This part — monitoring, sorting, filtering — is exactly what an AI agent does better than you, provided you give it access to the right data. Here's how to set up, in about an hour, a system where the agent monitors the market while you concentrate on what matters: choosing, applying, preparing for interviews.

Why an agent alone is not enough

First important clarification: a classic AI assistant, without connection to a data source, does not know the offers published yesterday. He reasons well, but on dated or generic information. For an agent to be useful to you in a job search, they need two things: reasoning skills (understanding your criteria, comparing offers, writing) and access to a truly up-to-date index of offers. This is the role of the MCP (Model Context Protocol), which allows an agent like Claude to be connected directly to an external database.

sourcing.sh thus presents an index of approximately 1.4 million job offers, continuously refreshed, accessible free of charge to job seekers via MCP, with free levels of 10, 100 or 150 requests per day depending on your usage. The agent queries the index; you're reading results, not job site pages.

Step 1: Plug your agent into the index

In Claude, add the MCP connector from sourcing.sh (the operation is done in the settings, without any particular technical skills). Once the connection is active, you can formulate your requests in natural language: “Find project manager offers published in the last three days in Île-de-France, excluding consulting firms. » The agent translates, queries the index, filters and presents you with a summary.

Step 2: define criteria that really filter

An agent is only as good as the brief he is given. Take the time to write, once and for all, a criteria record that you will paste at the start of the conversation:

• The position : accepted titles and related titles (the same profession is often published under 4 or 5 different names).
• Firm constraints : location or teleworking, salary range, type of contract.
• Exclusions : sectors, companies or keywords which indicate an offer to be rejected.
• Your differentiators : what, in your career, should guide the sorting (rare language, certification, sector of origin).

With this brief, ask the agent not for “offers”, but to sort them into three piles: to be processed today, to be looked at if time allows, to be ignored. It's this ranking that saves you time, not the raw volume.

Step 3: automate daily monitoring

The real value of the device is repetition. Every morning, the same request: “Which offers corresponding to my criteria record have been published since yesterday? » On a continuously refreshed index, this question has a precise answer — often 3 to 10 offers per day on a well-defined perimeter, a volume that you can actually process. Applying within 24 to 48 hours of publication significantly increases your chances of being seen, simply because you arrive before most of the queue.

A free level of 10 requests per day is more than enough for this daily ritual; The 100 and 150 query levels are used when you broaden the search or explore several avenues in parallel.

What the agent won't do for you

Let's be clear about the boundaries. The agent won't decide which company deserves your energy, won't interview, and shouldn't send out mass applications without your proofreading — a generic application is spotted, and paid for. The correct division of roles: the agent monitors, sorts and prepares; you choose, personalize and send.

Our conviction at sourcing.sh is that the tools of tomorrow will be databases queried by agents, rather than sites to consult. For a job seeker, this toggle is already available, and it is free: an index of fresh offers, your agent, your criteria. The market works continuously; your monitoring can now do the same.