# MCP (Model Context Protocol)

> MCP, the Model Context Protocol, is an open standard for connecting AI agents to external tools and data. In customer support it lets a user's own agent (Claude, OpenAI, or a custom one) call a product's capabilities as typed tools through a scoped endpoint, with each tool stating its effect before it runs.

Canonical: https://peeve.ai/glossary/mcp/

## What does MCP change for support?

It adds a second kind of user. Your product's next visitors won't all have eyes; some are agents acting for a person. MCP is how you serve them the same capabilities you serve humans (as structured tools with dry-run effects) instead of leaving them to scrape a chat widget built for people.

## Is MCP the same as an API?

No. An API is a surface you publish; MCP is a protocol an agent speaks, with tool discovery, typed inputs, and effect declarations built in. A product can expose its capabilities over MCP and a REST API at once. The MCP layer is what makes those capabilities legible to an autonomous agent.

**Related:** Capability registry, Agent experience, Prompt injection.