What is the Model Context Protocol?
MCP was released in November 2024 and has since been adopted across AI assistants, agent frameworks, and developer tools. Because it's open, anyone can implement it, whether they build the AI application or the system it connects to.
MCP is often compared to a universal port for AI. Before it, every pairing of an AI application and a business system needed its own custom integration. With MCP, a system describes its capabilities once, and any MCP-compatible AI application can discover and use them. This page covers the protocol itself. An MCP server is the software that exposes one specific system through it.
How does the Model Context Protocol work?
MCP defines three roles and a small set of things a server can offer. Messages between them use JSON-RPC, a lightweight standard for structured requests and responses.
| Part | What it is | Example |
|---|---|---|
| Host | The AI application a person uses | An AI assistant or coding tool |
| Client | The connector inside the host that talks to one server | One client per connected server |
| Server | Software that exposes a system's capabilities | A server for a data platform or file system |
| Tools | Actions the AI can call | Run a query, start a workflow |
| Resources | Data the AI can read for context | A file, a record, a schema |
| Prompts | Reusable templates a server offers | A standard way to request a report |
MCP servers can run locally alongside the AI application or remotely over HTTP. Remote servers typically use OAuth to confirm who is asking, so the AI only gets the access that person has.
What is an example of the Model Context Protocol in use?
One connection, many AI tools
A company wants its analysts to reach their data platform from whichever AI assistant they prefer. Instead of building a separate integration for each assistant, the platform publishes one MCP server.
Each analyst adds that server to their assistant and signs in. The assistant discovers the available tools, and the analyst can ask questions in plain language and get answers from the platform, within their own permissions.
Model Context Protocol vs. MCP server vs. API
| Model Context Protocol | MCP server | API | |
|---|---|---|---|
| What it is | An open standard | Software that implements the standard for one system | A system's programming interface |
| Who it is for | Anyone building AI connections | AI clients connecting to that system | Developers writing code |
| Analogy | The shape of the plug | A plug for one appliance | The appliance's internal wiring |
| Relationship | Defines how clients and servers talk | Often wraps an existing API | Can sit underneath an MCP server |
Why does the Model Context Protocol matter?
- Fewer custom integrations. One protocol replaces a separate connector for every AI application and system pair.
- Choice of AI tools. Teams can use the assistants they prefer while reaching the same business systems.
- Agents that can act. MCP gives agents a standard way to find and use tools, not just answer from memory.
- Governance still applies. With authorization built in, AI access can follow each person's existing permissions.
How does Alli do this?
Alli by PMG's MCP Server implements the Model Context Protocol for Alli. It reads Alli's API specifications, registers them as callable tools, and secures every call with OAuth 2.1 with PKCE, forwarding each user's authorized token so agents act within that person's access. Requests are stateless, so tokens aren't stored between calls.
Through it, MCP-compatible AI clients can reach more than 20 Alli services, including Datasources, Data Explorer, Generative Dashboards, Generative Workflows, and Creative Insights, and tool definitions stay in step with Alli's API as it changes.
Explore Alli MCP ServerModel Context Protocol FAQ
When was the Model Context Protocol introduced?
MCP was released as an open standard in November 2024. It is open for anyone to implement, and many AI applications and developer tools now support it.
What is the difference between MCP and an MCP server?
MCP is the protocol: the shared rules for how AI applications and systems talk. An MCP server is software that uses those rules to expose one specific system's tools and data.
Does MCP replace APIs?
No. MCP servers often wrap existing APIs and present them as tools an AI can discover and call. The API still does the work underneath.
Is MCP secure?
The protocol supports authorization, and remote servers commonly use OAuth so the AI acts with the signed-in user's permissions. Overall security depends on how each server is built and run.
Last updated ‹ Back to the glossary