What Are MCP Tools?
August 23, 2026 · 4 min read
MCP tools are individual functions that an MCP server exposes to AI agents. Each tool has a name, a description the AI reads, and a JSON schema describing its inputs. When an agent wants to do something — search your notes, create a file, send a message — it calls a tool.
Anatomy of a Tool
Every MCP tool has three parts:
Name — a machine-readable identifier (e.g. search_notes)
Description — a human/AI-readable explanation of what it does
Input schema — JSON Schema defining what parameters it accepts
The AI agent reads the description to decide when to use the tool, and uses the schema to construct valid arguments. No developer writes this integration — the agent figures it out from the metadata.
Tool Annotations
Tools can carry annotations — hints that tell the AI client how to handle them safely:
readOnlyHint
Tool only reads data — no confirmation needed
destructiveHint
Tool may destroy data — confirm before running
idempotentHint
Safe to retry — same result every time
xNotePadAI's 13 Tools
Here's what xNotePadAI's MCP server exposes, grouped by behaviour:
Read-only (no confirmation needed)
Write (safe, non-destructive)
Destructive (requires confirmation)
How Discovery Works
When an agent connects, it calls tools/list and receives the full catalogue:
// Agent sends:
{"jsonrpc":"2.0","id":1,"method":"tools/list"}
// Server returns 13 tools with schemas + annotations
// Agent now knows what it can doWhy This Matters
Traditional integrations are brittle — they break when APIs change. MCP tools are self-documenting. If xNotePadAI adds a 14th tool tomorrow, every connected agent discovers it automatically on the next tools/list call. Zero code changes on the client side.