MarcoPolo

Standing room only at the MCP Server Builder Series last night. What are all these developers curious about? How to actually build and run MCP servers in the real world. Here’s what the stellar lineup of speakers covered: 🔹 Aman Singla, my cofounder at Immersa.ai, spoke on curating enterprise context and delivering it to LLMs alongwith universal tools that these LLMs can invoke across 40+ SaaS apps, DBs, and data warehouses. Super charge your AI dev! 🔹 Meredith Hassett shared how Canva turned their developer experience into an agent experience in their docs — and used it to build their MCP server. 🔹 Michael Watson….🖥 Watson demo’d authentication in our community MCP server using WorkOS AuthKit inside the Apollo MCP server (and hosted the event 🙏). 🔹 Tobin South (WorkOS) gave updates from MCP Night 2.0. 🔹 David Emanuel Sarabia shared his personal story and showed how ClinicaMind is automating clinical workflows with voice AI. 🔹 Ravi Madabhushi demo’d how Scalekit secures a remote MCP server with drop-in OAuth. ⚡ The energy in the room said it all: MCP is real, and developers are shipping with it now. If you’re building agents or AI-powered apps, this is the community you want to be part of. 👉 Check out upcoming sessions here: https://lnkd.in/gZxegDtw #MCP #AIagents #LLM #ContextEngineering #OpenAI #DeveloperTools

Great energy at last night’s AI meetup hosted by AICamp in San Francisco! — the focus was on two crucial pieces of the AI puzzle: Agent Orchestration and Context to LLMs 🎯 While LLMs are getting smarter, the real value isn’t just in their pre-trained knowledge — it’s in how effectively you feed them the right “context”. The key lies in passing along the enterprise context that teams use daily to make informed decisions. 🚀 One standout was the launch of Model Context Repository (MCR) by Immersa, presented by Aman Singla. MCR acts like a version-controlled context brain for your AI stack — combining code, documentation, tools, and data into a continuously evolving memory system that: - Powers IDE copilots, chatbots, and internal agents - Connects what AI does to what your business knows - Enables reasoning, not just retrieval 📌 Think of it as git for AI context with the ability to grow as your teams use it over time. A big unlock for building reliable, enterprise-grade AI apps. Big thanks to the organizers and the Immersa team for a solid demo!

'Context is All you Need' 🚀 Attention AI Developers! We are thrilled to announce the launch of Model Context Repository (MCR) by Immersa at the AICamp event today in San Francisco 🌟 Join Us for an Exclusive Tech Talk Discover how to unlock the power of reasoning in large language models in the enterprise. In this session, Aman Singla will dive into practical strategies to deliver the right context to LLMs—helping them move beyond simple retrieval and into sophisticated reasoning. What you will learn: - How to structure and deliver contextual information for LLMs - Real-world methods for providing context across prompts and tools. - Leveraging the Model Context Repository (MCR) to build smarter, more capable AI apps 📅 Event Details Date: Today at 5.30 pm Location: AI Camp, San Francisco Sign up: See link in comments See you at AI Camp! #AICamp #ModelContextRepository #AI #LLMs #DeveloperCommunity #SanFrancisco #ContextEngineering

Here's why most AI projects stall after the initial successful prototype. Building an AI Agent is easy - connect a model, write some prompts, maybe throw in a RAG pipeline. It works. Demo approved. Stakeholders excited. Then the real problems start. The prototype falls apart in production: - Your RAG pipeline fetches stale or irrelevant data. - Your schema changes and suddenly all the answers are wrong. - Business logic lives in 8 different places — and none are versioned. - Devs are hard-coding context into prompts and hoping it doesn’t break. Context drift becomes real: the model says things that were true last week, but not anymore. Soon your “smart agent” is just a fancy lookup tool wrapped in retries. Why? Because context isn’t static. It changes when a table gets renamed. A config value is updated. A business rule changes. A human overrides a default. Most teams treat context as something you can embed once and forget. But in reality, LLMs need a live feed of your application state, business logic, and data lineage — not just a snapshot. This isn’t a prompt engineering problem. Even the term 'Context Engineering' doesn't do it justice. It’s an infrastructure problem. To scale AI Agents in production needs systems and frameworks to provide always current context. - Track context across data, code, APIs, and people. - Serve only the relevant slices to your LLM at inference time. - Keep context fresh automatically. - Handle edge cases, overrides, and uncertainty with auditability. Most developers will try to build this themselves. It works — until it doesn’t. We’ve been working with devellopers that are building AI copilots, customer agents, analytics agents, etc. What they all run into is the same thing: managing context becomes the bottleneck. This is the infrastructure no one talks about — but every serious AI team will need. If you’re building something with AI, I’d love to hear: How are you managing context today? What’s working — and what’s not? #AIEngineering #LLMInfrastructure #ContextEngineering #AIAgents #PromptEngineering #VectorSearch #DataInfrastructure #DevTools #AIOps

Building an AI App/Agents? Without context, your LLMs are just guessing. If you want your AI to be useful, you need to feed it the right context: 🧠 Entity resolution: user_id, cust_id, and account_id might be the same thing — but your agent won’t know that unless you tell it. 📄 Business logic: Pricing rules, approval workflows, SLA thresholds — none of this lives in the LLM by default. 🗂️ Semantic models: Raw schemas and table names don’t convey meaning. Agents need a layer that maps data to real-world concepts. 🧱 Tooling context: What APIs can the agent call? What functions can it invoke? What permissions does it have? 🕒 State and history: Agents need access to prior interactions, recent changes, and time-sensitive data to act intelligently. Context isn’t optional. It’s the difference between an AI Agent that works in prod and one that just looks good in a demo. Want help designing context-aware agents? Drop a comment or DM — happy to share how developers are using our Model Context Repository to solve these issues in production. #AI #LLM #AIagents #DeveloperTools #AIapps #PromptEngineering #SemanticLayer

A few weeks ago, I reached out cold to Jon Loyens at data.world for advice on a tough technical decision my cofounder and I were wrestling with. Jon didn’t know me. But he made time — simply because we’re building in the same space: making data easily consumable for AI. He walked us through his own founding journey, the tradeoffs he faced with a similar choice, and shared honest, thoughtful advice — no agenda, just a real conversation. A week later, I found out data.world was being acquired by ServiceNow. Anyone who’s been on either side of an acquisition knows how intense those final days are. That Jon still carved out time to help a fellow founder… I was already grateful. Now, I’m just in awe. 🔥 This is what makes the startup ecosystem special: founders helping founders, even when no one’s watching. Thank you, Jon — and congrats to you and the data.world team. ServiceNow couldn’t have made a better call.