Software Synthesis analyses the evolution of software companies in the age of AI - from how they're built and scaled, to how they go to market and create enduring value. You can reach me at akash@earlybird.com.
London AI Breakfast Series
July 2nd: MCP Ecosystem Breakfast
July 3rd: Value Selling for Fast Growing GTM Teams with Wiz, Databricks, Salesloft
July 4th: AI x Fintech Breakfast with Hummingbird
July 16th: Computer Use and Browser Agents
Ahead of this week’s MCP breakfast in London, I produced some slides (via Gamma) on the development of the protocol since its launch back in November (and since we first discussed it back in February).
What started as a project born out of David Soria Parra’s frustration while working with AI coding is now the universally adopted protocol for tool calling, embraced by OpenAI, Google and many of the leading Clients like Cursor, Cline, Zed and more.
To date, most of the usage has been by developers, as Jon Turow of Madrona has covered in his analysis of data from Smithery server downloads.
As one (very niche) joke goes, there are more people talking about developing MCP servers than people actually using them.
Looking across different registries like Smithery, MCP Market, PulseMCP, and MCP.so, there are thousands of servers by now.
Of these, the official MCP GitHub repo counts 275 official servers (PulseMCP counts over 500) - i.e. servers maintained by the vendor. These span a range of categories from observability vendors to browser automation and data infrastructure.
With the momentum the protocol has had, it was inevitable that forward-thinking software vendors would release their own official remote MCP servers to ready themselves for agent consumption.
The MCP team at Anthropic has been closely listening to the community and the most recent set of updates to the protocol are clearly geared towards making enterprise deployments viable.
On the authentication and security front, these releases will enable enterprises to connect their existing identity solutions to authenticate traffic to MCP servers.
Elicitation support will result in servers being able to solicit more input where needed to complete tasks, a prerequisite to more capable agents.
Structured outputs and better developer experiences will continue to drive more bottoms-up community momentum.
Anthropic teased some of these updates at their recent MCP Developers Summit and also shared examples of how enterprises are using MCP servers.
A Registries API is coming soon, which will likely include:
OAuth metadata for authentication flows
Capability descriptions for AI agents
Security verification for enterprise adoption
Multi-deployment model support (local/remote/enterprise)
It’s still very early days. Supabase’s inflection post-MCP with Lovable (h/t Sivesh) is one great example of the promise the protocol holds.
Soon enough, we’ll move on from the less inspiring examples listed here to ones like this, this, and more compelling use cases.
Data
Jobs
Early stage European AI companies in my network are actively looking for talent - if any roles sound like a fit for you or your friends, don’t hesitate to reach out to me.
If you’re exploring starting your own company or joining one, or just figuring out what’s next, I’d love to introduce you to some of the excellent talent in my network.
Reads
China's Evolving Industrial Policy for AI
Have any feedback? Email me at akash@earlybird.com.
That's a very well-researched article, thanks, Akash! I found the reference links useful as well 👍
Things are moving quickly in the MCP space: Madrona's article from April mentions 700k weekly downloads for TS SDK, we are at 5m now 😅