We appreciate that AI and the related terminology can be confusing. This article provides definitions of some key terms used by Mediafly when discussing generative AI and related concepts.
LLMs v Agents
LLM (Large Language Model)
An AI model that powers text generation and reasoning, such as Claude or GPT. LLMs generate responses based on patterns learned from large amounts of data, rather than executing fixed rules or formulas.
Agent / Agentic
An AI system that can take multi-step actions, such as retrieving data, applying logic, and generating a result, with minimal manual input at each step, rather than responding to a single, one-off prompt.
Command Center
Command Center is Mediafly's centralized, agent-first workspace for sellers and content admins. It surfaces relevant content, recommended actions and lets users complete them easily across the Mediafly platform using a combination of API integrations and MCP tools.
MCP (Model Context Protocol)
An open standard that allows an AI agent to connect to external tools and data sources in a structured, defined way. It works on top of APIs, the standard method software systems use to exchange data, but adds a consistent format that lets any AI agent understand what a tool does and how to use it, without needing custom-built logic for each connection. This is different from a traditional integration, which is typically built one-to-one between two specific systems. With MCP, a single tool can be connected to any MCP-compatible agent, and a single agent can work with any MCP-compatible tool, without rebuilding the connection each time.
MCP Server
The system that exposes a set of capabilities (tools, data, or actions) to an AI model using the MCP standard. Mediafly's current MCP server exposes tools for the Toolbuilder (Value) platform. Additional MCP tools, for other Mediafly capabilities, are in development.
MCP Tools
The individual functions or actions made available through an MCP server (for example, "create a business case," "retrieve a calculation," or "generate a report"). Tools are what an AI model actually calls to perform a task. Today, Mediafly's MCP tools are scoped to Value Toolbuilder (beta); more will be added as other capabilities are connected via MCP.
MCP Command
MCP Command is the name of Mediafly’s MCP server, where external agents such as Claude, Glean, Gemini will be able to connect to our growing suite of MCP tools across the Mediafly platform.
How Agents interact with MCP
An agent uses an LLM to decide which MCP tool to call, in what order, and with what inputs, based on the user's request. The agent doesn't perform the calculation or lookup itself; it hands that off to the appropriate tool, which returns a defined, structured result.
This is what distinguishes an MCP-connected agent from a general-purpose AI assistant: the agent's output is grounded in the tool's actual logic and data, rather than the LLM generating an answer from its own reasoning alone. As more MCP tools become available, an agent will be able to chain multiple tools together (for example, pulling company data, then generating a business case, then producing a report) to complete a broader workflow.
Prompts
The input or instruction a user gives to an AI model, whether typed as a question, a request, or a task description. The prompt is what the model responds to.
Turns
One exchange in a conversation with an AI model, typically a user's prompt and the model's response. A conversation is made up of multiple turns, and the model can use earlier turns as context for later ones.
One-shot prompting = a single turn to get the desired output
Skills
A packaged set of instructions and logic that directs an AI model to perform a specific, repeatable task in a consistent way (for example, generating a business case using a defined template and calculation method). Skills are designed to produce governed, predictable outputs, rather than open-ended results.
Determinism
A system is deterministic when the same input always produces the same output. Traditional calculation engines (like spreadsheet formulas) are deterministic. LLMs, by contrast, are probabilistic by default, meaning outputs can vary slightly between runs unless the system is specifically designed to constrain them.
Other Agentic System Terms
System Prompts
A set of instructions given to an AI model before the conversation starts, defining its role, behavior, and boundaries (for example, what task it should perform, what tone to use, or what it should not do). Users typically don't see the system prompt directly, but it shapes how the model responds throughout the conversation.
Prompt Injection
An attempt to manipulate an AI model's behavior by inserting hidden or misleading instructions into content the model processes (such as a document, webpage, or message), with the goal of overriding its intended instructions. This is a security consideration for any system that lets an AI model read external content or take actions.
Guardrails
Rules, checks, or constraints built into an AI system to keep its behavior within intended limits, for example, preventing it from taking unauthorized actions, generating inaccurate outputs, or deviating from its defined task. Guardrails are part of how a system stays governed and predictable.
Traces
A record of the steps an AI model took to produce a result, including which tools it called, what inputs it used, and what intermediate results it generated. Traces are typically used for debugging, auditing, or verifying how an output was reached.
Replies have been locked on this page!