MCP
The protocol that lets AI do work instead of describing it.
What does the MCP category cover?
What MCP (Model Context Protocol) is, how it differs from an API, and why it is the difference between AI that talks and AI that acts.
What MCP covers
MCP — the Model Context Protocol — is the layer that lets an AI model use tools rather than describe them. Instead of writing instructions for a human to follow, the model calls a defined action with defined inputs, gets a real result back, and continues from there.
This category explains the protocol without the jargon: what a tool definition is, why the boundary between reading and writing matters, how permissions and tenancy are enforced, and why an audit trail is not optional once software can change records on its own.
It also covers the honest limitations. MCP does not make a model smarter, it makes it capable. Capability without scoping is a risk, which is why most of these pages spend as much time on constraints as on possibilities.
What you'll learn
- What MCP is and the problem it was designed to solve
- Why read actions and write actions need different guardrails
- How tenancy and role scoping keep an agent inside its lane
- Why every agent action should be auditable after the fact
Pages in MCP
What Is MCP (Model Context Protocol)?
MCP is an open protocol that lets an AI model discover and call the tools of a software system — the difference between AI that talks and AI that acts.
ReadMCP vs API — What's Actually Different
APIs are built for developers who already know what to call. MCP is built for AI that has to discover what is possible and then act within rules.
ReadWhy APIs Return Data but MCP Executes Tasks
An API answers a precise question. MCP accepts an intent, chooses the operations, and carries a multi-step task to completion.
ReadCan AI Work Without MCP?
Yes — AI can answer, summarize and draft without MCP. What it cannot do without tool execution is finish work inside your business system.
ReadMCP Security
MCP security is about identity, scope, confirmation and audit. What must be true before an AI agent is allowed to write to your business data.
ReadMCP Servers Explained
An MCP server exposes a set of typed, permissioned tools that an AI client can discover and call. Here is what it contains and how a call flows.
ReadBuilding MCP Tools
Good MCP tools map to complete business actions, validate their inputs, fail loudly and stay idempotent. The design rules that separate usable tools from demos.
ReadHow AI Discovers MCP Capabilities
AI does not need to be trained on your software. It reads the tool catalog at connection time and selects actions from descriptions and schemas.
ReadMCP for Field Service Software
What MCP changes for contractors and field service businesses: an AI that can book, convert, invoice and follow up inside the real system of record.
ReadMCP server, client, tool, resource and transport
The five parts of any Model Context Protocol setup — server, client, tool, resource and transport — and why confusing them leads to false claims about what an AI agent can do.
ReadAuthentication vs authorization in MCP
Connecting an AI assistant to an MCP server proves identity. It does not grant permission. How authorization has to be re-evaluated on every tool call.
ReadOAuth MCP vs API-key MCP
OAuth MCP connects as a real user with revocable consent. API-key MCP connects as a shared machine credential. The operational and audit differences between them.
ReadMCP tool schemas and input validation
Tool schemas tell an assistant what arguments are legal. Server-side validation decides what actually runs. Why both layers are required.
ReadAdvertised capability vs registered handler
A manifest entry is a claim. A registered handler is code. A deployed handler that persisted a result is proof. Three different things routinely reported as one.
ReadTenant, role and entitlement enforcement in MCP
Three separate gates every agent tool call must pass: the right organization, a role that holds the capability, and an entitlement that includes the module.
ReadConfirmation tokens and two-step agent execution
How a preview-then-confirm pattern stops an AI agent from performing destructive, financial or outbound actions on a single ambiguous instruction.
ReadHuman versus agent execution
What an AI agent should execute, what stays with a person, and why an external assistant is a scoped consumer of a system rather than an operator of it.
ReadVerifying agent work and persisted results
A confident answer is not a completed action. How to verify that an AI agent's work persisted, and what an audit record has to contain to be evidence.
ReadMCP failure, retry and recovery
Agent calls time out, transports break, and assistants retry. How idempotency keys and typed errors stop a retry from becoming a duplicate invoice.
ReadThe MCP and agent infrastructure checklist
A practical checklist for evaluating any vendor's MCP or AI-agent claims: connection, discovery, permission, confirmation, execution and evidence.
ReadFrequently asked questions
How is MCP different from an API?
An API is designed for developers to call from code. MCP describes those same capabilities in a form a model can discover and invoke directly, including what each tool does and what inputs it needs.
Is it safe to let an AI agent write to my business data?
Only with three things in place: the agent is scoped to one organization, actions are limited by the acting user's role, and every call is logged with who asked, what ran and what changed.
Do I need to understand MCP to use AI in my business?
Not to use it, but it helps when comparing products. It is the difference between an assistant that tells you what to do and one that can do it.