The Doorway Between AI and Trusted Intelligence

The Model Context Protocol (MCP) is an open standard that defines a consistent way for AI applications to connect with external tools, data sources, and systems. Instead of building a custom integration for every combination of AI app and data source, developers use MCP as a shared interface that any compliant system understands. First released by Anthropic in November 2024 and later placed under the Linux Foundation's Agentic AI Foundation as a vendor-neutral standard, it has become a widely adopted way to connect AI to the real-world context it needs.

An MCP server is a program that exposes specific capabilities — such as access to a database, a file system, an API, or an analytics engine — to AI applications through the Model Context Protocol. It advertises what it can do in a standardized way, so any MCP-compatible AI client can discover and use those capabilities without custom integration code. Organizations build or deploy MCP servers to give AI safe, governed access to their own tools and data.

In MCP, the host is the AI application the user interacts with — such as a desktop assistant, an IDE copilot, or a custom agent — and the client is the connector inside that host which maintains a dedicated, one-to-one session with a single MCP server. A single host can maintain multiple client connections to different servers, giving the AI a unified set of tools and data across many sources. Together, the host and its clients handle discovering capabilities, sending requests, and routing responses.

MCP uses a client-server architecture built on JSON-RPC, a lightweight standard for exchanging structured messages. When an AI host connects to an MCP server, the two negotiate capabilities, and the server tells the client which tools and data it offers, so the AI can call those capabilities as needed during a task. Connections can run locally as a subprocess or remotely over HTTP, which lets the same approach work for both desktop apps and cloud services.

Before MCP, connecting AI applications to external tools and data required a separate custom integration for each pairing — a burden that grows quickly as the number of AI apps and data sources increases. MCP collapses this to N+M: build one MCP server for your database, and every MCP client can use it. This is why it's often compared to a universal connector like USB-C, replacing many one-off adapters with a single shared standard.

These are the three core capabilities an MCP server can expose. Tools are functions the model can invoke (such as database queries or API calls), resources are data sources the model can read, and prompts are pre-defined templates that guide workflows. By standardizing these building blocks, MCP lets any compatible AI application understand and use a server's capabilities consistently.

A traditional API defines how two specific software systems exchange data, and each API tends to have its own unique design that developers must learn and integrate individually. MCP sits at a higher level: it's a standard way to expose capabilities — which may themselves wrap underlying APIs — so AI applications can discover and use them uniformly. In short, APIs connect particular systems to each other, while MCP provides a consistent interface for AI to reach many tools and data sources at once.

Function calling (or tool use) is the underlying mechanism by which a language model decides to invoke an external function and supplies the arguments. MCP does not replace function calling — it standardizes it, defining a universal, discoverable way to describe tools, authenticate, and connect so the same tools work across different AI applications. Put simply, function calling is how a model triggers a tool, while MCP is how that tool is consistently exposed and made available to any compatible model.

Because MCP lets AI applications access real tools and data — and take actions — securing those connections is essential. Key considerations include strong authentication and authorization (modern MCP uses OAuth 2.1 authentication for remote servers), limiting each server's access to only what's necessary, and validating and monitoring the actions AI is permitted to perform. Treating MCP servers as governed entry points to sensitive systems, with clear access controls and auditing, helps keep AI use safe and compliant.

RAG focuses on retrieving information to feed into the model's prompt, while MCP enables both retrieval and action — the model can read data and write data through the same protocol. Framed simply, RAG makes a model more knowledgeable, whereas MCP makes it more capable. The two are complementary: an AI system can use MCP to connect to sources and still apply RAG techniques to ground its responses.

Qlik’s MCP server includes three lookup tools for Declarative Data Pipelines. Coding agents can search for spaces and connections on a Qlik tenant, get the details of a pipeline project, and search for the tables and views associated with a data connection. Data engineers can ask in natural language without leaving their IDE, which means less context switching during iterative pipeline coding.

Yes. Qlik MCP lets AI coding agents such as Claude Code and GitHub Copilot generate and update data pipelines from natural-language intent. The agents can also handle other data engineering tasks, such as creating data quality rules, managing data products, and delivering governed data for context-aware insight. It goes beyond pipelines, too: Qlik’s MCP server makes it easy to connect AI with the entire Qlik platform, so AI can reason with real business context wherever you work.