Model Context Protocol creates value by enabling user-driven extensible integrations
Soham Ganatra · He quit his job, went all-in on AI agents—then grew to 100K users & a $30M Series A in a year. | ... · PMF Show · Sep 10, 2025 · at 47:49
Soham Ganatra, co-founder of AI integration platform Composio, explains how the Model Context Protocol (MCP) differs from native developer-built application integrations.
0:00 / 1:00exact quote · 60.5s
▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark
“And so, MCP enables the second kind of integrations, which is like you could turn any sort of smart software into an extensible thing so that as a user, you could say, hey, I want my specific software like cursor to connect with Jira, right? Because Jira is what I care about. Or maybe Calendar to Cursor, because that's maybe what I care about, right? And so MCPs are great that way, because like, if every software becomes kind of like an MCP client, or like, maybe, let's say, good softwares like that you are using, maybe they become MCP client. You have this advantage of connecting that client to a server, which could be like by Notion, and you might say, hey, my, this software now connects to Notion, right? Because that's the integration I care about. So, so that, that's my take on it, which is like, it creates a lot of value by making every software into an extensible thing that like other, like users could decide what they want to connect it with.”
quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →
More from Soham Ganatra
Soham Ganatra: Current AI agents do not learn from repeated API errors
“Agents don't learn. So if my agent is connecting to Salesforce and Salesforce has like 10,000 different parameters and like it keeps messing up those parameters every time, it will keep doing that even like the third time, the fifth time, the 30th time, right?”
Soham Ganatra: Founders cannot validate AI product ideas by asking potential customers
“My hunch with AI space is that you can't validate problems by asking people because, like, they won't able, they won't get it.”
Soham Ganatra: Autonomous software integration agents are one LLM generation away
“I still think it's not a solvable problem today, maybe like one generation away.”
Charging independent developers is counterproductive when early product feedback is more valuable
“For me, at least the way I used to think about it was, like, a lot of these folks are, like, independent developers who are doing personal projects because they're, like, very interested in the new tech, and, like, to charge them right now is, like, kind of a …”
Foundation models and integration layers will eventually dismantle the SaaS application monopoly
“And like, so at some point, we can break into this sort of monopoly of, like, application layer, and we can say, like, you don't really need to care about applications. All you need to care about is essentially foundation models and, like, Composio, and, like,…”
Integration platform NPS drops drastically as users discover unsupported edge cases
“So, so at some point, like, it's almost, like, comical, but, like, your NPS is, like, 10, 10, 10, 10, 10, and then as, like, the time increases, your NPS drastically starts decreasing, because, like, everyone, like, starts discovering this and other things the…”