Protocol Quest

Tracks

1. 01🧱Why Protocols Matter

   0/5 lessons

   An LLM alone is brilliant and trapped

   Before any clever protocol, the question is simpler: why does the model need a contract with the outside world at all? This track names the isolation problem, walks the workarounds people tried first, and lands on the breakthrough that turned every modern agent into something buildable.

   Lesson list (5)

   • 01
     The Isolation Problem~22 min · llm, context-window, knowledge-cutoff
   • 02
     Before Tool Calling — The Workarounds~22 min · history, regex, react, scaffolding
   • 03
     The Tool-Calling Breakthrough~22 min · tool-calling, function-calling, structured-output
   • 04
     Protocol vs Library vs Convention~20 min · protocol, library, convention, spec
   • 05
     The Compatibility Promise~22 min · compatibility, rfc-2119, must-should-may, wire-format

   ✓Quiz · 5 questions→

2. 02🛠️Tool Calling Contracts

   0/5 lessons

   Schema, loop, side effects, evaluation

   Tool calling is four steps and four temptations. This track nails down the JSON Schema contract every modern model speaks, the agentic loop that chains many calls together, the discipline of side-effect boundaries, and the test-shaped habit that keeps your tools honest.

   Lesson list (5)

   • 01
     The Five-Step Flow~22 min · agentic-loop, round-trip, stop-reason
   • 02
     JSON Schema as Contract~26 min · json-schema, parameters, description, required
   • 03
     Tool Choice and the Agentic Loop~22 min · tool-choice, auto, any, loop
   • 04
     Side-Effect Boundaries~22 min · read-tool, write-tool, idempotency, human-in-the-loop
   • 05
     Tool Contract Tests~24 min · evaluation, regression, fixtures, ci

   ✓Quiz · 5 questions→

3. 03🏛️Provider Dialects

   0/5 lessons

   OpenAI, Anthropic, Gemini — three accents, one idea

   All three major providers speak the same tool-calling idea with subtly different envelopes. This track walks the OpenAI Responses API shape, Anthropic's three-tool-category split, Gemini's function declarations, and the comparison matrix that prevents 'why did this work last week' moments.

   Lesson list (5)

   • 01
     Three Houses, One Idea~22 min · openai, anthropic, gemini, compatibility
   • 02
     OpenAI Responses API~22 min · openai, responses-api, tool-calls, input
   • 03
     Anthropic Messages — Three Tool Categories~22 min · anthropic, claude, tool-categories, computer-use
   • 04
     Google Gemini Function Declarations~22 min · gemini, google, function-declaration, automatic-fc
   • 05
     Comparison Matrix~20 min · comparison, matrix, decision, portability

   ✓Quiz · 5 questions→

4. 04🧭MCP Architecture

   0/5 lessons

   From N×M chaos to one shared substrate

   The Model Context Protocol exists because every client was rebuilding every integration from scratch. This track names the N×M problem MCP solves, walks the Host-Client-Server triangle, and traces the initialize handshake that establishes capability before anything useful happens.

   Lesson list (5)

   • 01
     The N×M Integration Problem~22 min · nxm, integration, scale
   • 02
     Origin and Governance~18 min · origin, anthropic, spec, governance
   • 03
     Host, Client, Server~24 min · host, client, server, topology
   • 04
     Lifecycle and Capability Negotiation~22 min · initialize, lifecycle, capabilities, handshake
   • 05
     JSON-RPC 2.0 — The Wire~20 min · json-rpc, request, notification, error-codes

   ✓Quiz · 5 questions→

5. 05🧩MCP Primitives

   0/5 lessons

   Resources, Prompts, Tools — and the client-side three

   MCP defines six primitives, three offered by servers (Resources, Prompts, Tools) and three offered by clients (Sampling, Roots, Elicitation). This track maps the six, gives each a concrete shape, and shows where the lines fall when you are deciding 'should this be a Tool or a Resource?'

   Lesson list (5)

   • 01
     The Six-Primitive Map~18 min · primitives, overview, server-side, client-side
   • 02
     Resources — Read-Only Data~22 min · resources, uri, read, subscribe
   • 03
     Tools — Callable, Side-Effectful~22 min · tools, side-effects, schema, annotations
   • 04
     Prompts — Templated Workflows~18 min · prompts, templates, user-invokable
   • 05
     Sampling, Roots, Elicitation~22 min · sampling, roots, elicitation, client-features

   ✓Quiz · 5 questions→

6. 06🔌Transports

   0/5 lessons

   stdio for local, Streamable HTTP for the world

   Transports are how MCP messages cross the wire. Two are blessed: stdio for local subprocesses and Streamable HTTP for remote, scalable connections. This track walks both, plus the session, streaming, and resumability primitives that make the HTTP one production-ready.

   Lesson list (5)

   • 01
     stdio Transport~22 min · stdio, subprocess, local, framing
   • 02
     Streamable HTTP~24 min · http, sse, streaming, remote
   • 03
     Session IDs and Statefulness~20 min · session, mcp-session-id, stateful, stateless
   • 04
     Streaming and Resumability~22 min · streaming, resumability, last-event-id, reconnect
   • 05
     Choosing a Transport~18 min · decision, stdio-vs-http, operational

   ✓Quiz · 5 questions→

7. 07🛡️Authorization and Security

   0/5 lessons

   OAuth 2.1, origin safety, roots, human approval

   Protocols that touch real systems live or die on their security story. This track walks MCP's OAuth 2.1 framework, the origin / localhost traps, the role of Roots in scope-limiting servers, and the human-approval pattern that prevents the most expensive mistakes.

   Lesson list (5)

   • 01
     OAuth 2.1 — The Authorization Framework~26 min · oauth, resource-server, scopes, tokens
   • 02
     Origin and Localhost Traps~22 min · origin, localhost, dns-rebinding, cors
   • 03
     Root Directory Scope~18 min · roots, scope, filesystem, sandboxing
   • 04
     Human Approval and Audit Trails~22 min · approval, audit, destructive, logging
   • 05
     Threat Model~22 min · threat-model, prompt-injection, supply-chain, tokens

   ✓Quiz · 5 questions→

8. 08🧬Versioning and Compatibility

   0/4 lessons

   Revisions, feature detection, deprecation windows

   Compatibility is the unsexy half of every successful protocol. This track names MCP's date-based revision scheme, walks why feature detection beats version sniffing, and shows the deprecation window and compatibility-matrix discipline that keep servers and clients talking across years of releases.

   Lesson list (4)

   • 01
     Protocol Revisions~18 min · revisions, date-versioning, spec
   • 02
     Feature Detection over Version Sniffing~20 min · feature-detection, capabilities, sniffing
   • 03
     Deprecation Windows~20 min · deprecation, sunset, compat, windows
   • 04
     Compatibility Matrix~22 min · matrix, testing, ci, interop

   ✓Quiz · 4 questions→

9. 09🌐Ecosystem and Adjacent Contracts

   0/4 lessons

   MCP vs function calling vs REST vs webhooks

   MCP is one contract among several. This track puts it next to provider tool calling, classic REST/OpenAPI, and webhooks — naming where each shines and the heuristics for picking between them when designing a new integration.

   Lesson list (4)

   • 01
     MCP vs Function Calling vs REST~22 min · comparison, mcp, function-calling, rest
   • 02
     OpenAPI for Agents~22 min · openapi, oas, schema, agent-contracts
   • 03
     Webhooks and Async Events~22 min · webhooks, async, push, mcp-async-tasks
   • 04
     When to Use What~18 min · decision, heuristics, architecture

   ✓Quiz · 5 questions→

10. 10🚀Build a Server

    0/5 lessons

    From minimal MCP server to production

    All the theory in the world is no substitute for shipping one. This track walks the official Python SDK, the smallest useful server, Claude Desktop integration, observability, and the production shipping checklist that keeps your server upright after launch.

    Lesson list (5)

    • 01
      Minimal MCP Server (Python SDK)~22 min · python-sdk, fastmcp, minimal, scaffolding
    • 02
      Capability Manifest~18 min · manifest, capabilities, metadata, versioning
    • 03
      Connecting to Claude Desktop~22 min · claude-desktop, config, integration, debugging
    • 04
      Logging and Observability~22 min · logging, tracing, mcp-logging, metrics
    • 05
      Production Shipping Checklist~24 min · production, checklist, launch

    ✓Quiz · 5 questions→