Skip to content

mcp.md

Latest commit

 

History

History
67 lines (51 loc) · 2.32 KB

mcp.md

File metadata and controls

67 lines (51 loc) · 2.32 KB
Raw

MCP (Model Context Protocol)

What is MCP?

MCP is the open standard for connecting AI agents to tools and data sources. Think of it as "USB for AI" — a universal interface that lets any agent talk to any tool.

How MCP Works

┌──────────────┐     MCP Protocol     ┌──────────────┐
│   AI Agent   │ ◄──────────────────► │  MCP Server  │
│  (MCP Host)  │   JSON-RPC over      │  (Tool/Data) │
│              │   stdio or HTTP      │              │
└──────────────┘                      └──────────────┘

An MCP server exposes:

  • Tools — Functions the agent can call (e.g., read file, query database, send email)
  • Resources — Data the agent can read (e.g., file contents, database schemas)
  • Prompts — Pre-built prompt templates

Using MCP in multi-agent

Every agent definition can reference MCP tools:

# In a catalog agent definition
tools:
  - type: mcp
    server: filesystem          # Built-in MCP server
    description: Read and write files

  - type: mcp
    server: postgres            # Database MCP server
    config:
      connection_string: "${DATABASE_URL}"

  - type: mcp
    server: custom              # Your own MCP server
    url: "http://localhost:8080"

MCP vs Other Approaches

Approach Scope Standard? When to use
MCP Agent ↔ Tool Yes (Linux Foundation) Always — it's the universal standard
Function calling Agent ↔ Tool Provider-specific Simple single-model tools
A2A Agent ↔ Agent Yes (Linux Foundation) Multi-agent coordination
AG-UI Agent ↔ Frontend Growing Streaming UI updates

Popular MCP Servers

Server Purpose
filesystem Read/write local files
postgres Query PostgreSQL databases
github GitHub API operations
slack Slack messaging
browser Web browsing and scraping
memory Persistent memory storage

Resources