Test and Query Your GraphRAG Database with MCP Server

𝗦𝘁𝗼𝗽 𝗼𝘃𝗲𝗿𝘄𝗵𝗲𝗹𝗺𝗶𝗻𝗴 𝘆𝗼𝘂𝗿 𝗔𝗜 with massive API schemas! 🤯 See this new reShapr demo, where we tackle the challenge of MCP "Context Overload" using the GitHub GraphQL API. When you expose a massive schema to an LLM, the sheer volume of tools can lead to high token costs and "hallucinations." In this demo, we show how 𝗿𝗲𝗦𝗵𝗮𝗽𝗿 𝗴𝗶𝘃𝗲𝘀 𝘆𝗼𝘂 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗖𝗼𝗻𝘁𝗿𝗼𝗹: - 𝗙𝗶𝗹𝘁𝗲𝗿 𝘄𝗶𝘁𝗵 𝗣𝗿𝗲𝗰𝗶𝘀𝗶𝗼𝗻: Use the --io flag to import only the specific queries your agent needs (e.g., just the user query). - 𝗗𝗿𝗮𝘀𝘁𝗶𝗰 𝗡𝗼𝗶𝘀𝗲 𝗥𝗲𝗱𝘂𝗰𝘁𝗶𝗼𝗻: See the character count of the MCP tools list drop from thousands to a manageable, surgical set. - 𝗙𝗶𝗻𝗲-𝗧𝘂𝗻𝗲 𝘄𝗶𝘁𝗵 𝗖𝘂𝘀𝘁𝗼𝗺𝗧𝗼𝗼𝗹𝘀: Use YAML configurations to define exactly how the AI perceives your API surface. The goal? Moving 𝗳𝗿𝗼𝗺 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗢𝘃𝗲𝗿𝗹𝗼𝗮𝗱 𝘁𝗼 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗖𝗼𝗻𝘁𝗿𝗼𝗹 to make your AI agents faster, cheaper, and more reliable. 📺 Watch the demo in action here: https://lnkd.in/dUpYhuFq #MCP #AI #GraphQL #API #OpenSource GraphQL Foundation / The Guild Software / Agentic AI Foundation / The Linux Foundation 🙌

Three concepts everyone keeps mixing up: MCP, RAG, AI Agents. They're not alternatives. They sit at different layers, doing different jobs. MCP standardizes how an LLM talks to tools — GitHub, Postgres, your filesystem. No more custom glue code per integration. RAG controls what the model knows at runtime. Frozen weights, fresh context, citations included. Agents close the loop — plan, act, observe, repeat until done. A way to remember: — RAG reduces epistemic uncertainty. — MCP reduces integration complexity. — Agents reduce operational friction. You don't pick one. You compose them. The leverage comes from the stack, not the silver bullet. Compose, don't compare. #MCP #RAG #AIAgents #LLM #SoftwareArchitecture

I am sharing a new practical project I developed focused on building Artificial Intelligence applications: SQL and PDF RAG Creation Using LangChain. In this repository, I demonstrate how to build and orchestrate Retrieval-Augmented Generation (RAG) workflows by integrating LangChain, Groq, and Docker. Furthermore, this architecture was committed for embedding in a proprietary system. The project features two main approaches: 1) RAG with SQL Database: Using SQLAgent to enable the language model to query a PostgreSQL database directly. 2) RAG with PDF Files: Text vectorization with Ollama Embeddings and storage in Chroma DB for efficient semantic document search. The material includes a complete step-by-step guide for configuring the containerized environment and structuring the agents. It is an excellent starting point for anyone looking to explore the potential of LLMs connected to structured and unstructured databases. Feel free to check out the code and share your feedback. (Yes, the answer depends on the API used, and this is a low price. But, if you have enough tokens, you can optimize.) Repository: https://lnkd.in/dCRjiSpA #RAG #RetrievalAugmentedGeneration #ArtificialIntelligence #InteligenciaArtificial #LLM #GenerativeAI

I've been experimenting with this personally for quite some time, and today I'm excited to finally make it public. Introducing OCP — Open Context Protocol (v0.1.0) 🚀 https://lnkd.in/dMxpuT8P OCP is an open context layer built on top of MCP that adds persistent, retrievable context and coordination primitives for AI agent An open layer built on top of MCP that helps AI agents remember context across sessions. Website: https://lnkd.in/dMxpuT8P GitHub: https://lnkd.in/dg5SAezv PyPI: https://lnkd.in/dqPXmJPW Licensed under Apache-2.0 (code) and CC-BY 4.0 (specification). This is still an early v0.1.0 release, the foundations are there, but there's a huge amount of room to grow. I'd genuinely love feedback, ideas, and contributions from the community, whether that's embedding backends, storage adapters, integrations, spec discussions, or simply testing it and opening issues. If you've shipped agent memory in production: what's the one piece of context your agents keep losing? https://lnkd.in/dMxpuT8P #OpenSource #MCP #AIAgents #LLM #DeveloperTools #OpenStandards #OpenContextProtocol

The more tools you give your AI agent, the dumber it gets. Loading 5 MCP servers into Claude Code costs me 80,000 tokens before the agent does anything. By then, half its context is already spent — on schemas, descriptions, and tool names it might never call. It forgets earlier files. Re-asks things I already told it. MCP promised tools that were cheap, plentiful, discoverable. The moment you wire them up, you get the opposite. mcpx is my answer. Every server behind one CLI. The agent loads zero schemas upfront — it finds tools when it needs them, calls them on demand, and gets back to thinking. Agents shouldn't pay rent on tools they're not using. v1.6 ships today. github.com/codestz/mcpx #OpenCode #OpenSource #ClaudeCode #Codex #Anthropic #MCP #CLI #Gemini #Ollama

MCP solved the wire. Now solve the layer above. To connect an AI assistant to your calendar today, you install a server, edit a JSON file, and hope a background process stays alive. To connect a second tool, you do it again. We built models that reason across thousands of pages of legal text — and we ask their users to be sysadmins. That is the wrong layer. In December 2025, MCP became a Linux Foundation standard under the Agentic AI Foundation, co-founded by Anthropic, Block, and OpenAI. Industry consensus at the wire level is exactly the right outcome. Standards at the wire level should be standards. But the wire is the engine. Nobody buys an engine. Eight slides above walk through what sits on top: connections instead of servers, federated discovery, lazy schemas, an intent layer, and the browser as a universal fallback when no native integration exists. That is Ferridis — an opinionated reference design and Rust implementation for the experience layer above MCP. Designed in public. Apache 2.0. Full design at ferridis.k8x.uk What's the worst integration setup pain you've hit with an AI assistant in the last six months? #AI #MCP #OpenSource #DeveloperExperience #Architecture

Just shipped mcp-foundry, a hands-on TypeScript monorepo for learning Model Context Protocol (MCP) with Google Gemini. What it covers: → Building MCP servers with custom tool definitions → Connecting an AI agent host via stdio transport → Agentic tool-calling loops with Gemini → Real-world weather API integration with Zod validation MCP is becoming the standard way LLMs interact with external tools — and this repo breaks down exactly how that works under the hood. Open source 👇 https://lnkd.in/gKbWwaxE #MCP #ModelContextProtocol #Gemini #TypeScript #AIAgents #LLM #OpenSource

There are 𝐟𝐨𝐮𝐫 𝐬𝐭𝐚𝐠𝐞𝐬. Most stop at one. Only a few get to the last one. ➤ 𝐒𝐭𝐚𝐠𝐞 𝟏  "I know how to make API calls to the different LLM providers." ➤ 𝐒𝐭𝐚𝐠𝐞 𝟐 "I know how orchestration works and the benefits of agentic frameworks like LangGraph, Agent SDK, or the AI SDK." ➤ 𝐒𝐭𝐚𝐠𝐞 𝟑 "I know how to deploy LLMs on Kubernetes with vLLM, wire them to vector databases, and optimise the KV-cache with tools like LMCache. I also understand why reverse proxies like LiteLLM are necessary for production-ready applications" ➤ 𝐒𝐭𝐚𝐠𝐞 𝟒 "I think of agents as systems. Agentic systems 😎" - Stages 1–3 are tools. Stage 4 is a worldview. Luis Serrano and I are teaching Stage 4 today. 𝐍𝐨𝐭 𝐫𝐞𝐠𝐢𝐬𝐭𝐞𝐫𝐞𝐝? 𝐓𝐡𝐢𝐬 𝐢𝐬 𝐲𝐨𝐮𝐫 𝐥𝐚𝐬𝐭 𝐜𝐚𝐥𝐥 👇 🔗 https://lnkd.in/ewNqfXJc

Just uploaded a new eval - Formal Conjecture Bench - to hugging face. It contains 74 tasks of the form 'Prove the stated result in Lean' derived from the Google Deepmind Formal Conjectures repo. Formal Conjecture Bench was built using the Harbor evaluation harness created by the Terminal Bench team. It has a nice difficulty gradient with the easiest results highly accessible to weaker models and the hardest problems above the current frontier. https://lnkd.in/eGcFSHw7