Setup 2026-08-25 5 min read

Connect your AI to every EU tender in five minutes

One sentence to your assistant and it sets itself up. Here is what happens, and what to do if your client cannot configure itself.

Updated 25 August 2026

TED (Tenders Electronic Daily) is the online version of the 'Supplement to the Official Journal' of the EU, dedicated to European public procurement.

TED alone publishes roughly 1,500 open tender notices every working day, across every EU member state.

Say this to Claude, ChatGPT, or any assistant that can read a link:

Read https://victano.com/start.md and follow the onboarding instructions.

That is the install. The rest of this post explains what your assistant is doing, so you can help it along if your particular client needs a hand.

What that file tells it to do

start.md is written for an AI, not for you. It tells your assistant to check whether the service is open, add Victano as an MCP server, run the login exchange, and store the key. Four steps, and your assistant handles the ones it can.

Step one: adding the connection

MCP is the protocol that lets an assistant call outside tools. Some clients let the assistant configure itself; others only take connectors from their own settings. Your assistant will work out which it is in, but here is the honest split:

Claude Code can do it alone:

claude mcp add --transport http victano https://mcp.victano.com/mcp

Cursor, Windsurf, and other clients with a JSON config can also do it alone. The entry is:

{
  "mcpServers": {
    "victano": {
      "type": "http",
      "url": "https://mcp.victano.com/mcp"
    }
  }
}

The Claude desktop app takes the same entry, but the app only reads its config at startup, so you will need to quit and reopen it.

Claude on the web and ChatGPT cannot be configured by the assistant. You add the connector yourself: Settings, then Connectors, then Add custom connector, and paste https://mcp.victano.com/mcp. Leave the authentication field empty for now. Your assistant will tell you when it needs you to do this, then wait.

Notice there is no API key in any of that. You connect first and authenticate through the tools, which is the next step.

Step two: the login

Once connected, your assistant sees three tools and nothing else. It will ask for your work email, we send a six-digit code, and you paste the code back into the chat. Your assistant passes it on and the rest of the toolset appears.

The useful part: the conversation you are already in becomes authenticated. You do not disconnect, edit a config, and reconnect before your first search. You just carry on.

Your assistant also receives an API key. Where it can edit the client config, it will store the key itself so future conversations start connected. Where it cannot, it will show you the key once and tell you which field to paste it into. Save it when that happens, because we store only a hash and cannot show it to you again.

Step three: tell it about your company

Do this straight away. Matching against a vague profile returns vague results and you will conclude the product does not work.

Your assistant reads your website itself and asks you for the parts a website never says: which public contracts you have delivered in the last three years, which certifications you hold, roughly what your turnover is, which countries you can actually deliver in, and what you will not do. Then it stores that as your capability profile.

We never crawl your site. Your assistant does the reading and sends us what it found, which is also why you can correct it in conversation rather than through a form.

Ask it to run a match immediately and look at the top five. If you would not bid on any of them, say so and let it sharpen the profile. Two minutes there saves weeks of poor matches.

Step four: ask for things

The useful questions are the ones you would ask a colleague who had read everything:

• What should we bid on this week?
• Is this one worth bidding on, and what would sink us?
• What changed since I last looked?
• Read the specification and tell me what would disqualify us.

Your assistant also has access to a library of workflows: the weekly scan, the bid or no-bid decision, eligibility screening, clarification questions, the compliance matrix, the proposal skeleton. Ask it to list the workflows and follow one. They encode how bid professionals actually work, which is more useful than an assistant improvising a process.

If something goes wrong

Your assistant says the endpoint is not open. It checked our status and we told it the truth. Leave your email on the front page and we will say when it opens.

The code does not arrive. Check spam, then ask your assistant to send another. The code lasts fifteen minutes and you get five attempts.

It works, then stops in a new conversation. The key was not saved. Ask your assistant to add it to the client config as an Authorization: Bearer header, or paste it into the connector settings yourself.

Results look thin. Free accounts see data delayed by 24 hours and have daily limits. Your assistant will say when it hits one.

Anything else: [email protected].