AI Agents Engineering · Model Context Protocol · Lesson 2 of 3
Transports & function calling vs MCP
MCP is a client-server protocol (JSON-RPC under the hood). A host app runs one or more MCP clients, each connected to an MCP server. Two standard transports carry the messages.
- stdio: the client launches the server as a local subprocess and talks over stdin/stdout. Best for local tools (filesystem, local DB).
- Streamable HTTP: the server runs as a remote HTTP service (supersedes the older HTTP+SSE transport). Best for hosted / multi-user servers.
Function calling vs MCP
- Function calling is the model-level mechanism: how one app tells one model about its tools.
- MCP is the integration standard: how tools/data are exposed so any MCP-aware client (Claude apps, IDEs, agents) can discover and use them.
- They compose: an MCP client typically surfaces MCP tools to the model as function-calling tools. MCP does not replace function calling — it standardizes where the tools come from.
Not a competitor to function callingA common mistake is framing MCP vs function calling as either/or. Function calling is the model's tool-use API; MCP is a transport/standard for delivering tools to that API across apps.
◆ Lock it in
- Transports: stdio (local subprocess) and streamable HTTP (remote service).
- Function calling = per-model tool mechanism; MCP = cross-app tool/data standard.
- MCP tools are surfaced to the model as function-calling tools — they compose.
Feynman drill — say it out loudExplain MCP to someone who already understands function calling — what does it add, and when would you use stdio vs streamable HTTP?
Step 1 rate your confidence · Step 2 pick your answer
Which statement is correct about MCP transports?
Step 1 — how sure are you?