← Blog

Monitoring EU Tenders with an AI Agent: TED + OpenAI Agent Builder

One of the busiest production workloads on Pipeworx is EU procurement monitoring built in OpenAI Agent Builder. Here's the pattern, step by step — TED tender search as an MCP tool, no code and no API key.

The EU publishes thousands of public procurement notices every working day on TED (Tenders Electronic Daily) — the official journal for every contract above the EU publication threshold. Consultancies, suppliers, and bid teams monitor it for a living: new calls for tenders in their CPV codes, award notices naming competitors, deadlines that can’t be missed.

That workflow turns out to be a near-perfect AI-agent use case — and we can see it happening. One of the busiest production workloads on the Pipeworx gateway right now is EU tender monitoring running inside OpenAI Agent Builder: an agent that searches TED on demand, filters by country, value, and procurement category, and summarizes what changed.

Here’s the whole pattern, reproducible in about five minutes.

TED’s own portal supports saved searches and email alerts. What it can’t do:

  • Answer questions. “Which German municipalities opened cloud-infrastructure tenders over €1M this month, and what are the deadlines?” is a query, not a filter.
  • Cross-reference. An agent can pull the tender, then check the buyer’s history, the incumbent’s award notices, or currency conversions in the same conversation.
  • Summarize on your terms. Notices are long and formulaic; an agent extracts buyer, value, deadline, and fit in two lines.

The tools

Pipeworx exposes TED through two MCP tools (the ted-eu pack — anonymous, no key):

  • search_notices — expert search across all EU procurement notices. Filters: free-text query, country (ISO alpha-3: DEU, FRA, ITA…), cpv (8-digit Common Procurement Vocabulary code), date_from/date_to, value_min/value_max (EUR), and notice_type (cn-standard for open calls, can-standard for awards, pin for prior information).
  • get_notice — one notice in full by publication number.

Updates land daily, matching TED’s publication cycle.

Build it in Agent Builder

  1. Open Agent Builder and create a workflow.
  2. Click the Agent node → Tools → + → MCP Server → + Server.
  3. URL: https://gateway.pipeworx.io/pipeworx-catalog/mcp — label pipeworx, authentication None.
  4. In the Agent node’s Instructions, tell it what to watch:
You monitor EU public procurement for [your sector]. When asked what's new,
call the pipeworx TED tools: search_notices with notice_type "cn-standard",
the relevant CPV code, and a date_from of the last check. For each hit,
report buyer, country, estimated value, deadline, and a one-line summary.
Use get_notice for details. Cite publication numbers.
  1. Preview with: “What IT-services tenders (CPV 72000000) opened in Germany in the last two weeks over €500k?”

The agent calls search_notices with cpv: "72000000", country: "DEU", notice_type: "cn-standard", value_min: 500000 — and returns cited publication numbers you can verify on TED directly.

Publish the workflow to ChatGPT and the bid team gets it as a chat; wire the same workflow to the Responses API and it becomes a scheduled digest.

Extending the pattern

The same shape covers the rest of the procurement world through the same connection:

  • UK: the uk-contracts pack (Contracts Finder) — search_notices, recent_notices.
  • US federal: usaspending for awarded contracts, federal-register for upcoming rule changes that move procurement.
  • Context: FRED/ECB rates for currency and indexation clauses, company packs for buyer and competitor lookups.

That’s the point of a gateway: the tender-monitoring agent that starts with TED doesn’t need a new integration when the question crosses a border.

Try it