Thousands of public tenders are published every day. For companies bidding on public contracts, the problem is almost never finding one tender: it’s finding the right one, buried in a constant stream of publications scattered across dozens of portals, thousands of different CPV codes, and contracting authorities describing the same subject matter in a thousand different ways.
With Tender Search, the new feature in Tender Brain, an AI agent does this work for you — and it does it by speaking your language, not by filling in faceted search forms.
What we built
Behind Tender Search is an MCP server (Model Context Protocol) that we developed in-house: a connector that gives Tender Brain’s AI agents direct, structured access to every tender published on official e-procurement registries, starting with ANAC data.
In practice, the agent doesn’t “browse” a website the way a person would: it queries a structured database directly, cross-referencing in real time filters on free text, CPV codes, place of execution (down to the NUTS/ISTAT detail of the individual lot, not just the contracting authority’s registered address), amount, and procedure status.
The difference from a text search on a portal is substantial: a traditional search engine returns everything containing a keyword. An agent connected via MCP understands the request, chooses the right queries to cross-reference, and — crucially — filters out false positives, a skill that normally requires an experienced eye for the sector.
A real case
To see what changes in practice, here’s a search performed with Tender Brain for a cleaning company, PuliTutto Spa, looking for active tenders in Puglia worth at least €200,000/year.
The request
“I run a cleaning company, I’m looking for all tenders in my field in Puglia worth at least €200K/year.”
A request written exactly as you’d phrase it to a colleague, with no need to know CPV codes or query syntax.
What the agent did
The agent cross-referenced several searches on the tender registry (free text “pulizia”/“servizi di pulizia”, location Puglia, minimum amount, active status, sector CPV codes such as 90910000-9 and related ones), then manually validated each result against three criteria:
- The lot’s actual location (not just the contracting authority’s registered office, which can be misleading with centralized procurement bodies covering multiple territories);
- CPV code genuinely consistent with building cleaning services (excluding plumbing works, portable toilets, or cemetery services that mention “cleaning” only in the description);
- Economic value recalculated on an annual basis, dividing the total amount by the contract’s actual duration — a decisive step when tenders span multiple years.
The result
Alongside the tender that fully matched the target, the agent also flagged a similar but below-threshold tender (Comune di Giovinazzo, ≈ €138,010/year, below the €200K requirement) and discarded, with reasons, three tenders that mentioned “cleaning” in their description but actually concerned watercourse regulation works, portable toilets for events, and cemetery services with cleaning as an ancillary component.
That’s the point: it’s not enough to find tenders containing a keyword. You need to understand what that tender actually is, and Tender Search delivers the reasoning, not just the list.
Why natural language, and why MCP
The MCP (Model Context Protocol) is the standard that lets an AI agent connect to external data sources in a structured, secure way, instead of having to “improvise” by scraping a website. For Tender Brain, this means giving the agent direct, reliable access to the universe of published tenders, with the same ease as querying the documentation of a single tender already uploaded to the tool today.
For anyone working in a tender office, the result is that the scouting phase — today often made up of email alerts, saved searches on portals, and manual checks on dozens of bids a day — compresses into a request written the way you’d ask an experienced colleague, with the same ability to tell a real opportunity from a false positive.
Try Tender Search
Tender Search is available to Tender Brain customers on the Pro and Business plans, integrated into the same environment where you already manage the documentation for the tenders you bid on.
Book a free demo and discover how to find your company’s next right tender, without spending hours on portals.