The MCP server vs custom agent framing in section 1 is the most underappreciated point here. Everyone defaults to building a full agent when half the time an MCP server that exposes your product's API to existing agents would ship faster and get used more. The 34% dashboards stat from your MCP server basically proves this. Curious whether you've seen MCP usage grow faster than the built-in agent over the last few months.
We've only started tracking the usage in-app between the two recently, but they seem similar so far. MCP has had a lot more search traffic and people going to the product page though.
The MCP server vs custom agent framing in section 1 is the most underappreciated point here. Everyone defaults to building a full agent when half the time an MCP server that exposes your product's API to existing agents would ship faster and get used more. The 34% dashboards stat from your MCP server basically proves this. Curious whether you've seen MCP usage grow faster than the built-in agent over the last few months.
We've only started tracking the usage in-app between the two recently, but they seem similar so far. MCP has had a lot more search traffic and people going to the product page though.
Good post , thanks for sharing your experience! I have some Q : how did your load balancer handle spikes in traffic when transferring data to Kafka?
Or your team implemented sharding ?