What is an MCP server?
The server does three jobs. It publishes a list of tools, describes what each one does and what it needs, and handles the calls when an agent uses them.
Before MCP, connecting an AI assistant to a business system meant building a custom connector for each pairing: an HTTP client, authentication, documentation, and code to maintain. MCP replaces that with one shared protocol. A system exposes its capabilities once, and any MCP-compatible AI client can discover and use them.
How does an MCP server work?
An MCP server sits between an AI client, such as an assistant or an agent framework, and the system it connects to. The client asks the server what tools are available, the server describes them, and the agent decides which ones to call to complete a request.
- Connect. A person adds the MCP server to their AI client and signs in, so the server knows who is asking.
- Discover. The agent asks which tools are available. The server returns only the tools that person is allowed to use.
- Request. The person asks for something in plain language, such as "check which data connections failed this week."
- Call and chain. The agent calls one or more tools in sequence. The server runs each call against the real system, with the person's authorization.
- Review. The agent returns the result, and the person decides what to do with it.
What is an example of an MCP server?
Checking data connections from an AI assistant
A marketing analyst works in an AI assistant all day. Instead of logging in to a separate platform, they ask the assistant to list any data connections that stopped syncing and compare the schema of one source with last month's.
Through an MCP server, the assistant discovers the right tools, calls them with the analyst's permissions, and returns the answer in the conversation. No one wrote custom integration code to make that possible.
MCP server vs. API vs. custom connector
An MCP server usually wraps an existing API rather than replacing it. The difference is who it's designed for: people writing code, or AI agents choosing tools.
| Custom connector | API | MCP server | |
|---|---|---|---|
| Built for | One specific integration | Developers writing code | AI assistants and agents |
| How tools are found | Hard-coded | Read in documentation | Discovered automatically by the agent |
| Reuse across AI clients | None | Each client integrates separately | Any MCP-compatible client |
| Maintenance when the system changes | Update each connector | Update client code | Tool definitions update with the system |
| Permissions | Depends on the build | Depends on the build | Can forward each user's own authorization |
Why do MCP servers matter for business teams?
AI assistants are only useful at work if they can reach the systems where work happens. MCP servers make that practical and governed at the same time.
- AI where people already work. Teams use the assistant they prefer, and it reaches business systems directly.
- Less integration work. When a team adds or switches AI assistants, nothing needs to be rebuilt on the system side.
- Governed access. Agents can act within each person's existing permissions, so access rules still apply.
- Agent-ready data. Clean, governed, clearly described data gives an agent less room to pick the wrong tool or misread a field.
How does Alli do this?
Alli by PMG's MCP Server is a Model Context Protocol server for Alli. It reads Alli's API specifications, registers them as callable tools, and secures every call with OAuth 2.1 with PKCE. It forwards each user's authorized token with every request, and requests are stateless, so tokens aren't stored between calls.
Through it, an AI agent can work across 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 ServerMCP server FAQ
What does MCP stand for?
MCP stands for Model Context Protocol, an open standard for connecting AI models to tools and data. An MCP server is the piece that exposes a system's tools through that protocol.
Is an MCP server the same as an API?
No. An MCP server usually wraps an existing API and presents it as tools an AI agent can discover and call. An API is built for developers writing code. An MCP server is built for AI agents that pick which tools to call.
Do you need to write code to use an MCP server?
Usually not. Once an MCP server is added to an MCP-compatible AI client and the person signs in, they can make requests in plain language and the agent calls the tools.
Are MCP servers secure?
They can be. A well-built MCP server authenticates each user, typically with OAuth, and lets agents act only within that user's existing permissions. Security depends on how the server handles authorization and tokens.
Can an MCP server take actions, or only read data?
Both are possible. An MCP server can expose read tools, like querying data, and action tools, like starting a workflow. What an agent can do depends on the tools exposed and the user's permissions.
Last updated ‹ Back to the glossary