Prospecting precision-farming centres
- First working version
- 2 h
Goal
An agent that finds Europe’s leading precision-farming centres, plots them on a map and tracks down the work email of the people in charge.
How it works
First the user hands it the criteria: which countries, what kind of centre and which lines of work matter. The agent chains searches across the web, and for each candidate it does not stop at the name: it goes into the site, reads what they specialise in, what trials they publish and who they work with, and writes one line on why that centre matters. Anything that misses is dropped right there, with the reason noted. When a site will not be read directly, it opens it in the browser through the Chrome MCP and looks at it.
Then comes finding the person. For every centre that passes the filter it works through the staff page, the project PDFs and the press releases until it identifies who owns that line of work, and from there gets their work address. Several agents run at once, one per block of centres.
No address is taken on trust: each one is checked against a second source and comes out flagged as verified, inferred from the domain pattern, or generic. None is invented.
To finish, it plots the centres on a map and pours everything into a spreadsheet, so who gets written to can be decided without searching for anything again.
- Deliverable
- An HTML map with the centres plotted and a record for each one.
- A spreadsheet with centre, country, what they specialise in and why they match the criteria.
- The person in charge at each centre with their job title, their address and how certain that address is.
- Stack
- Claude Cowork
- Claude web search
- Subagents in parallel
- Chrome (MCP)
- HTML map
- Excel skill
Why it matters
Finding the contacts, verifying them and understanding the context around each one are enormously expensive in time, and no company can do without them. The luck of it is that most of that work can be automated.
What could come next
- 01Run the addresses through Apollo before handing the list over. Right now an inferred address is delivered as inferred, and someone has to decide whether to risk it; checking it against a contact database turns that label into a yes or a no, and the one part of the deliverable that needs a second look goes away.
- 02Wire it into the ERP, so the contacts land straight where customers and opportunities are already managed. Today they end up in a spreadsheet someone has to copy across by hand, and that copying is where fields get lost, companies get duplicated and nobody records who spoke to whom.
Worth knowing
- legal
Looking up the names, job titles and email addresses of specific researchers is processing personal data. These are professional details published by the institutions themselves, but they should be used only for the contact you actually intend, not stockpiled into a list that outlives the campaign.
- future
Contacts expire. In this campaign one department head had retired, one researcher had left his centre and another organisation had changed domain. What a website publishes is not always what is true today.
- dependency
When a site has to be checked in the browser, the agent works on a session that is already open: it does not sign in for you and it does not get past verifications. If a site asks for a login or a captcha, it stops there and notes it.