"Building MCP Server Tools with Zod Schema Validation"

MCP tools extend AI superpowers and give to us huge opportunities to enhance our agents. But here’s the hidden cost: every server adds context, even when idle. Every unused MCP is burn our precious tokens. That's why better to disable MCP, which we don't use often or use tools, as `hypertool-mcp` to load only the essentials. After weekly clean up unused MCP, you’ll see the token savings immediately.

How do you go from a blank project to your first fully tested AI agent? In our latest blog, we share how we adapted Test-Driven Development (TDD) into the LLM era, using evals to define desired behaviors before writing code. By supercharging Claude Code with Model Context Protocol (MCP) servers, we grounded the agent in the right context (docs, repos, wikis).  ✔️ Faster setup of evals with Eval Protocol ✔️ Richer, more reliable test suites generated automatically ✔️ A safer workflow for iterating on prompts, models, and features without regressions This shift turns AI into a true development partner- expanding tests, validating behaviors, and helping engineers scale agent capabilities with confidence. 👉 Read the full blog here: https://lnkd.in/gQX5aN-6

To people wrapping the entire world with MCP: if a tool already has a good, self-documented command-line interface, you probably DON'T need to turn it into an MCP server. If the CLI of a tool makes it difficult for an agent to use it, please do not write an MCP interface to make it smoother for AI agents. Write a *better* command-line interface to make it smoother for AI agents *and* humans *and* scripts. If something can just run locally and serialize its state to disk, it should be a regular CLI tool, simple as that. Thus, example of a questionnable MCP use-case: https://gitmcp.io/

Built an MCP-powered workflow in Eigent AI that automates GitHub tasks using a Personal Access Token (PAT) for secure authentication. The agent fetches user profiles, repositories, pull requests, and generates structured Markdown reports and summaries efficiently...

Friday – Quick wins (Pro Tips) for MLOps 🎯 Ready for fast impact with your MLOps strategy? Here are your quick-win moves this week: Pro Tips: Choose a single use case and pilot a full CI/CD pipeline for it. Set up real-time monitoring dashboards—drift, performance, fairness. Automate a retraining trigger—e.g. when drift > threshold or KPI dips. 📖 Read more: 🔗 https://lnkd.in/gPUvEa9i 🔗 https://lnkd.in/g8xp6wKr Here is a practical example: 🔗 https://lnkd.in/gdz7wh89 👉 Next week, we're diving into “Designing Responsible AI Systems”—from fairness-by-design to explainability and ethical guardrails. See you then! #MLOpsWins #MLEngineering #NextWeekPreview

🤯 MCP is going crazy viral right now! 𝐔𝐒𝐁-𝐂 𝐦𝐨𝐦𝐞𝐧𝐭 𝐟𝐨𝐫 𝐀𝐈. Anthropic proposed "Let's all use the same JSON format when connecting AI to tools" and everyone said "Sure." In this video on "What is Model Context Protocol" by Gurleen Kaur, explains what changed Before and After MCP. 👉 Before MCP: - LLMs were heading in different directions - Every integration was custom with M apps × N tools = M×N integrations - Engineers needed a common language 👉 After MCP: - Build one MCP server for your tool - M + N integrations instead of M×N - Works with any MCP-compatible AI app AI apps can now instantly connect to any tool or live data. If you can write JSON, you are the new MCP GOAT. Building some complex Multi-Agentic Systems? Let's build together with CosX.

Behind the shine of LLMs is an operational iceberg: vector databases, agent memory, serving & inference, execution/tooling, and observability with p95 metrics. That’s how AI moves from slides to the P&L. #AIAgents #AIStack #LLM #RAG #VectorDatabase #AgentMemory #Serving #Inference #Tooling #Observability #MLOps #LangGraph #LangChain #LlamaIndex #AIEngineering #ProductionAI

Agents do not fail loudly when tools drift. MCP helps. Problem Custom tool glue gets brittle as endpoints multiply. What works - Standardize tool discovery, auth, and errors with MCP - Keep tools as small HTTP services and version them like any API - Document capabilities in one place and test locally the same way CI does - Add an evaluation gate so bad tool changes never reach prod Result Fewer silent-wrong runs. Faster local setup. Safer deploys. Takeaway / Question Are you standardizing agent tools with MCP, or rebuilding glue for each feature? #AI #Agents #MCP #RubyOnRails #APIDesign #DevTools #SoftwareEngineering

Machine Learning and Operations, in short MLOps. The life cycle of MLOps is an iterative and never-ending process. The most important steps in the life cycle are data versioning and parameters, and result registration.

We’ve launched our MCP server 🚀 fileAI has powered mission-critical file workflows for finance and operations teams from banks to factory floors for years. Now we’re making those same AI OCR, classification, and schema capabilities available where adoption often begins: the developer terminal. By embedding fileAI directly into MCP-compatible environments, the same reliability trusted by enterprises is now in every developer’s toolkit. Whether you're a business leader or developer, see how you can automate your files on our fresh site: https://lnkd.in/ecEsqJh3