The Integration Problem

N applications times M tools

Without a standard, connecting N AI applications to M tools or data sources means writing roughly N × M custom integrations, each with its own authentication, schema format and error handling. Every new assistant must re-implement the same GitHub, database and calendar connectors; every tool vendor must support every assistant. A shared protocol turns this into N + M: each tool is wrapped once as a server that speaks the protocol, and each application implements the client side once. This is the same idea that made USB, HTTP and the Language Server Protocol successful. Standards also make security review and tooling (inspectors, gateways, registries) possible in one place.

Tools down, agents across

MCP standardises how an agent reaches tools and data; A2A standardises how agents talk to other agents.

One charger for every phone

Before a common connector, every device came with its own plug. A shared standard lets any device use any charger.

N x M versus N + M, run

I ran this with plain Python 3 (standard library only). With 8 applications and 25 tools, custom glue needs 200 integrations; a shared protocol needs 33 (8 clients plus 25 servers).

apps, tools = 8, 25
print("custom glue  :", apps * tools, "integrations")
print("shared protocol:", apps + tools, "(", apps, "clients +", tools, "servers )")

Output:

custom glue  : 200 integrations
shared protocol: 33 ( 8 clients + 25 servers )

Quick check: What does a shared protocol change about integrations?

• It makes models smaller
• It removes the need for tools
• From N x M custom connectors to N + M implementations
• It replaces authentication