Model Context Protocol (MCP)

Key points

• MCP is an open standard (created by Anthropic) for connecting agents to tools and data.
• Think 'USB-C port for AI tools' — one uniform interface instead of custom glue everywhere.
• An agent (the client) talks to MCP servers that expose tools and resources.

The one-line definition

Model Context Protocol (MCP) is an open standard that defines a common way for AI agents to talk to tools and data sources. An MCP client (your agent) connects to MCP servers that expose tools (actions) and resources (data), all through the same interface — no custom integration per tool.

Note: MCP standardises tool access. Build the connector once; reuse it everywhere.

The problem MCP solves

Without a standard, every tool you connect (GitHub, your database, Slack, Google Drive) needs its own hand-written integration. With N agents and M tools, that's roughly N × M custom connectors to build and maintain — it explodes. MCP turns this into N + M: each tool exposes one MCP server, each agent is one MCP client, and they all speak the same protocol.

Before MCP vs after MCP

BEFORE MCP (custom glue, N x M) AFTER MCP (one standard, N + M) ────────────────────────────── ───────────────────────────────

Agent A ──custom──► GitHub Agent A ─┐ Agent A ──custom──► DB Agent B ─┤ Agent A ──custom──► Slack │ speak MCP Agent B ──custom──► GitHub ▼ Agent B ──custom──► DB ┌────────────────────────┐ Agent B ──custom──► Slack │ MCP protocol │ ...tangled mess 🌀... └───────────┬────────────┘ │ one connector each ┌──────────┼──────────┐ ▼ ▼ ▼ GitHub DB Slack (server) (server) (server) ✅

How a client talks to MCP servers

┌──────────────────────────┐ │ 🧠 AGENT (MCP CLIENT) │ └───────────┬──────────────┘ │ 1. 'What can you do?' (discover) ▼ ┌────────────────────┐ │ MCP PROTOCOL │ uniform request/response └─────────┬──────────┘ ┌─────────┼──────────┐ ▼ ▼ ▼ ┌──────────┐ ┌────────┐ ┌──────────┐ │ MCP SRV │ │MCP SRV │ │ MCP SRV │ │ GitHub │ │ DB │ │ Filesystem│ └────┬─────┘ └───┬────┘ └────┬─────┘ │ exposes: │ exposes: │ exposes: ▼ ▼ ▼ tools: create tools: query resources: _issue, get_pr _rows files, dirs

1. Agent calls a tool ──► 3. Server runs it ──► 4. Returns result

A tiny code example (read it like English)

From the agent's side, MCP makes tools feel uniform: connect to a server, ask what tools it offers, then call one by name. The agent doesn't care if it's GitHub or a database underneath — the protocol is the same.

# The agent (MCP client) connects to an MCP server
client = MCPClient()
client.connect("github-server")

# 1. Discover what the server can do (no custom code per tool)
tools = client.list_tools()
# -> ['create_issue', 'get_pull_request', 'search_code', ...]

# 2. Call a tool by name through the standard interface
result = client.call_tool(
    "create_issue",
    {"repo": "my-app", "title": "Bug in login"},
)

# Same client, same calls would work for a DB or Slack server
print(result)

When does MCP earn its keep?

MCP mistakes beginners make

Remember these lines

• MCP = USB-C for AI tools: one standard plug instead of custom glue per tool.
• Agent = client; the tool/data provider = server exposing tools and resources.
• MCP standardises tool ACCESS; RAG grounds ANSWERS in retrieved text — different jobs.

Key takeaways

• MCP is an open standard (by Anthropic) for connecting agents to tools and data through one uniform interface.
• Think 'USB-C port for AI tools': build a connector once, reuse it across agents and tools.
• An MCP client (the agent) talks to MCP servers that expose tools (actions) and resources (data).
• MCP standardises tool access, turning an N×M integration mess into a clean N+M setup.

Frequently Asked Questions

Related topics

• Goal Setting & Monitoring
• Learning & Adaptation
• Memory Management Pattern
• Multi-Agent Collaboration
• Prompt Chaining Pattern

Browse all Agentic AI Patterns topics →