The Yellow Pages for AI: how smart tools find each other
Earlier we told how AI agents can communicate with each other via MCP and A2A. But talking alone is not enough. They must first find each other.
And that's exactly where directories come into the picture.
Think of it as a modern-day Yellow Pages, but for AI: a central catalog where agents search for other agents with the right skills. Internally in your organization, or just on the open market.
Without directory, you work with static links. With directory? Then you open the door to a dynamic network in which agents start smart collaborations in real-time.
From static to smart: this makes directories indispensable
Without directory, you're stuck with manual links. A developer then has to program in advance exactly who his agent can talk to. That works fine for simple tasks, but it limits you. Agents can't search for better help themselves or flexibly respond to new situations.
Directories make collaboration dynamic. Your AI agent can find the right "colleague" in the moment - internal or external - based on skills, context or location. That makes your platform scalable, smart and future-proof.
This is how AI agents find each other through a directory
The process of finding and collaborating follows a few logical steps, with the directory acting as the matchmaker.
Step 1: Publication (issuing the calling card).
A developer has built a new agent, for example, a "Valuta agent. That same developer creates an Agent Card for his agent: a JSON file with name, description, skills (such as currency_convert) and the agent's unique online address (an endpoint).
Next, the developer submits the Agent Card to a relevant A2A directory. This can be a public directory (for public services), or an internal one (for internal business tasks).
Step 2: Search & Find (locating the right specialist).
Another agent -- for example, a "Travel Agent" -- plans a trip to Japan, and needs to convert a price in Euros to Yen. It itself has no currency conversion function, so it does a search in the directory: "Wanted: agent with skill currency_convert and support for JPY." The directory searches the database and returns the Agent Card of the Currency agent.
What if there are multiple agents offering the same skill?
Then the searching agent itself selects the best match - based on filters such as performance, specialization, availability, or even AI-generated recommendations.
Just like you'd rather choose the best plumber than just anyone from the phone book, the agent chooses the one best suited for the job.
Step 3: Direct connection (the phone call).
From this point on, the directory is no longer involved. Communication is now direct between the two agents.
With the digital business card in hand, the Travel agent contacts the Valuta agent. He uses the candle's address to send a direct A2A request with the instruction, "Convert €150 from EUR to JPY."
Step 4: Execution & response (getting the job done).
The Currency Agent receives the order and retrieves the latest exchange rate. It converts the amount and sends the result directly back to the Travel Agent. The Travel Agent can now use this information in its own process.
Where do you find AI directories?
Because this technology is so new, there is not yet one central place like a "Google" or "App Store" for AI agents. The infrastructure is currently being built.
Still, you can divide directories into three categories:
1. Public & experimental directories.
These are public websites where developers and pioneers can register their agents to test the technology. They are intended as a testing ground to see how an open ecosystem of agents might work.
The best-known example is: A2ARegistry.org. This is an independent, community-driven project that serves as a public "Yellow Pages." Developers can publish their Agent Cards here, and other developers can use an API to conduct targeted searches for agents with specific skills.
Good to know: These directories are intended for experimentation, not (yet) for critical enterprise applications.
2. Private & proprietary directories
This is where most of the development is happening right now and where you'll see it most often in practice. Large organizations are building their own, internal directory accessible only within the corporate network.
Why?
- Security: A company like ABN AMRO or Philips does not want their internal "invoice-control agent" or "production-planning agent" to be findable on the public Internet.
- Control: The company can control which agents are allowed into the directory, so they can be sure they are reliable and secure.
- Privacy: Company-sensitive processes and data remain completely within the organization's walls.
How it works: A company sets up an A2A registry on its own servers. All AI agents developed within the company are registered in it. For example, if the marketing agent needs the sales agent, he does not look for him on the Internet, but in the company's secure, internal "yellow pages.
3. Specialized directories
It is expected that in the future there will not be one giant, central directory for everything, but rather multiple, specialized directories for specific sectors.
Consider:
- Financial directories: A registry specifically for fintech and banking agents, pre-screened for financial regulation and reliability.
- Medical directories: A heavily secured directory for healthcare agents, which comply with all privacy laws (AVG).
- Logistics directories: An open directory for agents in the transportation and supply chain industry to work together to optimize freight routes.
Summary: directories make your AI smarter, more flexible and scalable
- Want to experiment? Check out public projects like A2ARegistry.org
- Are you using it seriously? Then work with private, internal directories within your organization
- Looking ahead? Then you see a landscape emerging of specialized directories by sector
Important: An AI agent does not "find" a directory by itself. It is configured to use a specific registry as a starting point to discover other agents.
Want to know more about the basics? MCP (Model Context Protocol) is a standard for linking AI models to external data and systems. A2A (Agent2Agent) is an open protocol that allows AI agents to securely communicate and collaborate with each other within complex workflows. Read the extensive blog on MCP & A2A.