// Track

Building Production MCP Servers

From Beginner MCP to Servers You Run 24/7

The gap between a demo MCP server and one you trust at 3 AM is six engineering dimensions. Build the full production stack: HTTP transport, bearer token auth, scoped access, structured observability, read/write safety gates, and the audit lab — modeled on Knox's semantic memory layer (12 tools, 414 tests, 92% coverage, running 24/7 on a persistent server).

Recommended: Complete Skills, Hooks & Slash Commands first

9 lessons~82 min total

Lessons are shown in recommended order. Complete them in sequence for the best experience — or jump to any lesson.

Lesson 398·9 min read

Beyond Beginner MCP — What 'Production' Means

The gap between a demo MCP server and one you trust at 3 AM is six dimensions — reliability, auth, observability, transport, error handling, and stable output contracts.

Lesson 399·8 min read

The Three Primitives in Depth — Tools, Resources, Prompts

Tools do work, resources are data, prompts shape conversations — but the decision tree for which to use is more nuanced than it sounds, and most production servers get it wrong.

Lesson 400·9 min read

Designing the Tool Surface — Inputs, Outputs & Consumer Contracts

Grep your callers before you define your schema. The output shape serves the consumer, not the implementation — and breaking changes get a new tool name, not a version bump.

Lesson 401·8 min read

Transport & Lifecycle — stdio vs HTTP/SSE, Startup, Health, Reconnection

The transport choice is the single most consequential production decision in MCP server design — stdio servers die when the terminal closes; HTTP servers run until you tell them to stop.

Lesson 402·9 min read

Auth & Secrets in MCP Servers — Bearer Tokens, Scoped Access, Never Leaking

The model reads tool output. Any secret in that output is a leaked secret. Bearer tokens, scoped access per caller, and five leak patterns you must never implement.

Lesson 403·9 min read

Read vs Write Safety — The Permission Model, Destructive-Tool Gating, Idempotency

A destructive tool that fires automatically is a liability, not a feature. Three safety tiers, explicit confirmation gates, and two idempotency patterns that make write tools safe to retry.

Lesson 404·10 min read

Building a Knowledge-OS Bridge — A Real MCP Server

12 tools, namespaces, query/remember/recall — the exact architecture of Semantic Memory Layer, 414 tests, 92% coverage, running 24/7 on a persistent server.

Lesson 405·10 min read

Observability & Failure Modes — Logging, Split-Brain, Health Proofs

Heartbeat is not pipeline health. The split-brain incident where SO_REUSEPORT let two servers share a port — and neither knew the other existed — is the most dangerous silent failure in MCP operations.

Lesson 406·10 min read

The Production MCP Playbook — Packaging, Registration, Deployment, Audit Lab

Package → register in ~/.claude.json → deploy on the production server → run the 16-check audit lab. The exact playbook Knox used to ship Semantic Memory Layer to production in May 2025.