7 Comments
User's avatar
JohnFurrier.substack.com's avatar

Love this post. Disruptive enablement is what protocols do.

Expand full comment
Jon Turow's avatar

John, thanks so much! Looking forward to catching up soon.

Expand full comment
Haley Rae Derge's avatar

Thank you for such an in-depth exploration of the MCP ecosystem! It's fascinating to see how MCP is changing the landscape for AI agents beyond just chat interactions. The comparison to Coca-Cola's journey with infrastructure really highlights the importance of synchronized development between applications and infrastructure.

Expand full comment
Jon Turow's avatar

@haley thank you so much for reading! It’s indeed amazing how much history rhymes, from Coca Cola through to ChatGPT

Expand full comment
Nitin Badjatia's avatar

This is the best explanation of the MCP developments that I've seen, even for the non-technical crowd.

Expand full comment
Nimish Sanghi's avatar

Nice explanation.

I wonder if the point about MCP solving the impedance mismatch between LLM's probabilistic output and structured need of apis is the key benefit as mentioned in the example "without MCP" and "with MCP".

That problem is solved by tool calling capability of LLM which is induced during instruction tuning while training LLMs. MCP based agents eventually also uses this capability to decide which MCP service (tool) to call and what parameter values to provide.

Rather MCP helps loose couple the actual tool implementations from the use, making MxN problem into M+N. Once a tool/service is wrapped into an MCP server, anyone can use it in a standard way instead of coding the whole thing up himself/herself.

Expand full comment
ITNAmatter's avatar

Great thought piece! If you are interested in collaborating on building MCPs, reach out to me at jonathan@jeantechnologies. We are building MCP tools for businesses as well.

Expand full comment