All concepts
MCP Protocol
Expose tools and resources to AI clients through a standardized protocol.
Agentic AI · Advanced · ~8 min
In plain English
A standard plug for connecting tools and data to an AI application, so the same server works with any client that speaks it.
Why it's worth your time
It turns every integration from bespoke glue into something reusable — write the connector once, use it everywhere.
If you remember three things
- Server exposes tools, resources and prompts; client consumes them
- One integration, many hosts
- Tool descriptions from a server land in your prompt — treat them as untrusted
Overview
Model Context Protocol is an open standard that lets AI clients discover and call external tools, resources, and prompts through one typed contract. A server advertises its capabilities; any compatible client invokes them the same way, making integrations portable instead of bespoke per app.
How it works
- Start: AI Client An agent or chat app needs access to external tools and data.
- AI Client -> MCP Server The server advertises tools, resources, and prompts using a standard contract.
- MCP Server -> Tool Call The client invokes a tool with typed inputs.
- Tool Call -> Resource Files, docs, or database records are fetched with permissions.
- Resource -> Result Standardization makes tools portable across compatible clients.
In an interview
MCP is a standardized interface between AI clients and external tools or data. A server exposes tools, resources, and prompts with typed schemas; the client discovers them and invokes tools with structured inputs. The point is portability — write a server once and any MCP-compatible client can use it, replacing N-by-M custom integrations.
Production defaults
- Server design
- same rules as any tool surface: verbs, few tools, actionable errors
- Trust
- a third-party server's tool descriptions are text you didn't write, entering your prompt. Review them
- Scope
- least-privilege credentials per server. A connector is a permission boundary
What breaks
- Too many tools once several servers are connected — The count is cumulative. Enable per task, not globally.
- A connector's description changed your agent's behaviour — Expected — descriptions are prompt text. Pin versions and review updates.