Artificial Intelligence is only as good as the data it draws from. As enterprises explore ways to connect their trusted information to AI systems, a new standard known as the Model Context Protocol (MCP) is paving the way.
At Schema App, we’ve taken that next step by launching our own MCP Server, designed to make our customers’ Content Knowledge Graphs directly usable by AI assistants and copilots.
If you read our recent article, What is Model Context Protocol (MCP) and How Does Structured Data Play a Role?, this new article builds on that foundation to explain how we’re putting MCP into action.
What the Schema App MCP Server Does
Artificial Intelligence is only as good as the data it’s allowed to see.
Schema App’s Model Context Protocol (MCP) Server changes that by securely exposing your organization’s Content Knowledge Graph to AI assistants and copilots, such as ChatGPT, Microsoft Copilot, Claude, Gemini, Perplexity, and any other emerging MCP-enabled tools.
Think of it like a USB-C port for AI: a standardized, reliable way to plug high-quality, governed data directly into modern AI systems without custom engineering.
The Schema App MCP Server was purpose-built to turn your Content Knowledge Graph into a reliable AI data layer. It does three things exceptionally well:
- Delivers governed answers—not guesses.
It exposes only the structured, approved facts from your Knowledge Graph, so AI tools stop hallucinating and start answering with confidence.
- Acts as a secure gateway between your enterprise data and AI assistants.
Authentication, permissioning, and rate-limited access ensure your proprietary information is protected while still being usable in real-time conversations.
- Provides a consistent schema and ontology foundation for AI.
Because everything is backed by the same structured data layer used across Schema App’s platform, AI outputs stay aligned with your taxonomy, terminology, brand language, and definitions.
Internally, we’re already using the MCP Server to power copilots that reference Schema App’s own website content and product documentation. This helps us observe how different AI models interpret structured data, measure grounding accuracy, and identify new metrics and features to deliver to customers. These real-world experiments directly shape how the MCP Server evolves.
By making your structured data accessible through MCP, your AI tools can deliver:
- Accurate, brand-aligned answers grounded in your verified content.
- Traceable, compliant responses backed by structured data.
- Seamless integration with MCP-enabled tools with no custom connectors required.
Why We Built Our MCP Server
Enterprises are eager to utilize AI to enhance search, customer support, and content management operations. However, without a governed and structured data layer, AI systems risk hallucinating or misrepresenting your brand.
The Schema App MCP Server was built to solve that problem. It provides a safe, standards-based way to connect the structured data you already manage through Schema Markup and your Content Knowledge Graph directly to AI systems.
Unlike retrieval-augmented generation (RAG), which can pull from unverified sources, MCP ensures precision, traceability, and compliance by serving authoritative information directly from your knowledge graph.
How It Works
Connecting your Content Knowledge Graph to AI tools is a simple, secure three-step flow:
- Point the client — In your MCP-compatible client, add Schema App’s endpoint:
https://mcp.schemaapp.com/mcp.
- Authenticate — Sign in with your Schema App credentials; the platform enforces access controls, permissions, and rate limits, ensuring only authorized agents can query your graph.
- Query your data — Once connected, agents can query your graph in real time with no custom connectors or extra scope configuration, returning answers grounded in your governed content.
Prerequisites & notes: You need a Schema App Content Knowledge Graph (or robust site Schema Markup) and an MCP-compatible client (toolkits/clients from Microsoft, OpenAI, Google and others). Schema App’s Editor/Highlighter and Entity Hub can build a production-ready content knowledge graph, and our MCP implementation emphasizes governance and traceability to reduce hallucinations and support compliance.
Availability: The MCP Server is currently available to enterprise customers for internal use — contact your Customer Success Manager or use the Get Started flow to enable it.
Real-World Use Cases for Model Context Protocol (MCP)
Schema App’s MCP Server is already being tested across several enterprise applications, including:
-
- Healthcare: Benefits-coverage and clinical-info Q&A for patients and staff (prenatal care, plan coverage, clinical trials). Find a Specialty Physicians or Clinics in my city.
- B2B Tech: Product capability summaries and industry-specific success stories used in sales and marketing collateral.
- Customer Success: Onboarding and training lookups powered by company knowledge graphs.
- Finance / HR / Compliance: Policy, benefits and finance Q&A where answers must be authoritative, versioned and auditable.
- E-Commerce: Product Q&A, availability checks and guided shopping experiences driven by structured product and catalog data
Internally, our team is also experimenting with MCP to connect Schema App’s own website data to ChatGPT and Copilot tools. This helps us understand how AI interprets structured content and what metrics we can report on through our MCP Server—insights that will benefit our customers as we expand the feature.
The Results of MCP So Far
Early pilots and independent research both show that grounding LLMs with a governed knowledge graph delivers big, measurable gains in accuracy, trust and speed. Third-party benchmarks are dramatic — for example, KG grounding hit 91% accuracy vs 43% for GPT-4 on a clinical QA benchmark, and LinkedIn reported +78% accuracy / −29% resolution time when adding a knowledge graph to customer-service flows.
Our NLWeb + MCP demos and customer pilots mirror those results. Agents return more accurate, on-brand answers with explicit graph citations (reducing hallucinations) and teams iterate faster because the same governed data layer serves SEO, marketing, and AI. We also log the graph paths and citations used by agents, which improves explainability and auditability in production scenarios.
Practically, we measure grounding/citation rate, accuracy, hallucination rate, time-to-resolution, and end-user satisfaction in pilots — and those KPIs consistently improve as customers move from RAG-only systems to an MCP + Knowledge Graph architecture.
Benchmarks / third-party studies
- John Snow Labs: Clinical Q&A showed KG grounding produced 91% accuracy vs 43% for GPT-4 alone — a clear example of how structured knowledge changes outcomes in sensitive domains.
- LinkedIn (SIGIR ’24): Adding a Knowledge Graph to RAG/assistant flows produced +78% accuracy and −29% resolution time in customer-service bots.
- Microsoft GraphRAG and academic benchmarks: Multiple studies show KG-grounding improves factuality, explainability and multi-hop reasoning versus baseline RAG or LLM-only approaches.
- Data.world: KGs provide 300% Increase in Accuracy for LLM Responses in Enterprises.
What’s Next for Model Context Protocol?
We’re focused on turning MCP from a secure connector into a full production pipeline for AI—adding analytics and performance metrics, richer observability (logged graph paths, citations and reasoning chains), and deeper integrations with standards like NLWeb so teams can ship AI-driven search and copilots with confidence. NLWeb + MCP is already the turnkey path we use for branded chat assistants, and our observability work makes the data exchanges inspectable and auditable in production.
Our goal is simple: make your Schema App data not just machine-readable, but AI-ready—accurate, traceable and brand-controlled across every AI channel. MCP is already being exercised in pilots and internal demos (our product, engineering, CSM and sales-engineering teams run NLWeb/MCP experiments), and the Server is available to enterprise customers for internal team use.
If you’d like to enable MCP for your organization’s internal teams or explore NLWeb+MCP for a branded assistant, contact your Customer Success Manager to get started.
Mark van Berkel is the co-founder and COO of Hunch Manifest and the creator of Schema App. He is an expert in Semantic Technology and Semantic Search Marketing. Mark built Schema App to solve his own challenges in writing and validating schema markup.