PulseMCP

One of the challenges of building a communication protocol like MCP: the ecosystem faces a repeated "chicken or egg" problem with its long tail of rich features. Servers don't implement features until clients support them, and vice-versa. Sometimes the problem is even worse: servers can't adopt features lest they break (popular) laggard clients. For example, if a server wants to use dynamic tool updates to implement a multi-step `search_for_tools` then `execute_tools` pattern; you can't do that without breaking all the clients that don't yet support dynamic tool updates. They'd be stuck looking at your search_for_tools tool with no way to get the updated list. We partnered with Jan Čurn, CEO of Apify to write about how Apify works around this problem with their MCP server work: a community-powered effort called `mcp-client-capabilities`. https://lnkd.in/gTeZKvEj

Matt Martin, CEO of Clockwise, takes us through the difference between data access and intelligent coordination when building MCP servers. Most MCP servers expose calendar data for LLMs to process, Clockwise's server does the scheduling optimization server-side and returns actionable proposals. One leverages the LLM's intelligence. The other leverages domain intelligence. Read it here: https://lnkd.in/gyVUPPzb Appreciate the shout-out in DevToolsDigest, Heavybit!

Today we introduce one of our most ambitious project so far - Skybridge. An open-source framework to design ChatGPT Apps. ChatGPT Apps unlock a powerful new interface for interactive UI-driven chat experiences. But developers have been struggling with complex state sync, dual interaction surfaces, and missing frontend tooling. Skybridge solves this. It provides: ✅ A React library providing hooks, components, and the runtime glue to render your widgets inside ChatGPT's iframe environment. ✅ A drop-in MCP SDK with type-safe widget registration ✅ A Vite plugin and dev loop with HMR and other improvements https://lnkd.in/ewqS6k3p

MCP gateways help solve aggregation but often contribute to tool overload. When you have 1000+ tools available, agents struggle and context windows explode. When a design team just needs a few tools across Figma, Linear, and Github MCP Servers, they are likely to end up unintentionally loading hundreds. Our friends at MintMCP partnered with us to share how they are tackling this problem with Virtual MCP Servers. Give it a read here: https://lnkd.in/gHgPDnFJ

On a demo call today, I was asked if Model Context Protocol (MCP) would still be here in a year from now. I get it: the space moves so quickly, that it can be hard to identify where to invest efforts into. That said, I am betting that MCP is here to stay and here's why: For AI to do real work for us, it will need access to our data. This includes proprietary data sources such as our emails, customer channels, data warehouses, or internal research. We need two things for this to happen: (1) a way to securely connect to these systems with guardrails and access controls, and (2) a way for the AI to interface with the data systems to pull the data - tools or function calls are most natural here: functions are a core part of computer systems. MCP creates a standard for performing (1) auth and (2) tool specs. On top of that, it's been rapidly adopted across both AI clients and app providers. My take is that we will always need a solution that solves these two problems and MCP is now the established protocol for us to agree on how to do it. In a year, the protocol will change, and new approaches might come up - but these two core problems will never go away. I expect that MCP will look vastly different a year from now as more innovations come in for making tool calls for effective (code-based tool calls), better user experiences (MCP-UI), and better security/auth (oauth flows). Whether we call it MCP or something else, the AI world is converging on a shared protocol layer between AI and data: and that layer is here to stay. Michael Grinich Ido Salomon David Soria Parra Theodora Chu Tadas Antanavicius Toby Padilla Shannon Williams

I’m excited to share that the Official Model Context Protocol (MCP) Registry API just hit an API freeze (v0.1)! 🚀 Over the next month, the v0.1 API will stay stable and free of breaking changes, so integrators can start building on something solid. The goal of this freeze is to give developers a stable target for integrations and also to collect feedback from all of you so we can refine the API as we move toward v1 for GA! So if you're testing or integrating, reach out on Discord or GitHub, we would love to hear from you! A huge thank you to everyone for your feedback, ideas, contributions and of course for your patience through the fast-paced updates of the past several weeks. Your involvement has been key to getting us here and is exactly what a community-driven project looks like! ❤️ 🔗 https://lnkd.in/dsWGyMvm Massive shout-out to my fellow maintainers - Tadas Antanavicius, Toby Padilla, Adam Jones for driving this forward together! #mcp #modelcontextprotocol #opensource #oss #api #community #registry #developerexperience #devex #engineering #ai

https://www.pulsemcp.com/ by Tadas Antanavicius and Mike Coughlin are awesome weekly reads worth your time for keeping up to speed on `deep AI`