Coding Agents in Practice (3/5) — Building MCP Servers: Spec, Examples, Debugging

AI ์ฝ”๋”ฉ ์—์ด์ „ํŠธ ์‹ค์ „ (3/5) — MCP ์„œ๋ฒ„ ๋งŒ๋“ค๊ธฐ: ์ŠคํŽ™·์˜ˆ์ œ·๋””๋ฒ„๊น…

Cursor, Claude Code, and VS Code Agent all adopted the Model Context Protocol. Build a server yourself to understand it.

ํ•ต์‹ฌ ์š”์•ฝ

  • MCP is a standard protocol for LLM agents to access external tools, resources, and prompts
  • Primary sources: modelcontextprotocol.io spec, Anthropic Claude Code MCP docs
  • Three core capabilities: Tools (functions the agent calls) / Resources (read-only data) / Prompts (reusable templates)
  • Debugging is mcp inspector + stderr logging — that's the de facto standard
  • Pitfall: a single stray print() to stdout can break the entire JSON-RPC channel

1. The problem MCP solves

Each agent had its own tool-integration mechanism: Claude Code had one tool spec, Cursor had its own extensions, VS Code had a separate API. The same integration had to be built three times.

MCP standardizes this: write one MCP server, and it works in every MCP-supporting agent (Claude Code, Cursor, VS Code Agent, Continue, etc.). As of 2026 it is effectively the default.

Protocol essentials: JSON-RPC 2.0 + three capabilities (Tools / Resources / Prompts) + three transports (stdio / SSE / Streamable HTTP).

2. The three capabilities

2.1 Tools — Functions the agent calls

The most-used capability. The agent takes action. - Examples: search_database(query), create_jira_ticket(title, body), run_shell(cmd). - Definition: name + JSON Schema for arguments + short description. - Returns text or structured data.

2.2 Resources — Read-only data

Resources the agent reads but does not change. - Examples: file contents, DB rows, configuration values, API responses. - Identified by URI: file:///path, db://table/row/123, config://app/setting. - Unlike tools, no side effects are guaranteed (safe for agents to pre-load).

2.3 Prompts — Reusable templates

Frequent task patterns exposed as prompt templates. - Examples: "review this code," "explain this SQL," "draft a PR message for this diff." - Can be triggered as slash commands.

3. Minimum example — A Python MCP server in 30 lines

Using the mcp Python SDK. This is a real, working MCP server:

from mcp.server.fastmcp import FastMCP

mcp = FastMCP("hello-mcp")

@mcp.tool()
def add(a: int, b: int) -> int:
    """Add two integers and return the sum."""
    return a + b

@mcp.tool()
def greet(name: str) -> str:
    """Greet a person by name."""
    return f"Hello, {name}!"

@mcp.resource("config://app/version")
def app_version() -> str:
    """Read-only application version."""
    return "1.0.0"

if __name__ == "__main__":
    mcp.run(transport="stdio")

Register with Claude Code:

claude mcp add hello-mcp --command "python /abs/path/server.py"

That single command makes add, greet, and config://app/version auto-discovered inside the session.

4. Debugging — Looking inside an invisible channel

MCP's biggest pitfall is that stdout is the protocol channel. Under stdio transport, JSON-RPC messages flow through stdout — adding print("debug") immediately corrupts the channel.

Four debugging rules: 1. Never write to stdout. Logs go to sys.stderr or a file. 2. Use the MCP Inspector — npx @modelcontextprotocol/inspector python server.py opens a GUI to invoke tools/resources/prompts and see responses. 3. Raise errors. Don't swallow with try/except — raising propagates the error to the client. 4. Validate JSON Schema. The SDK auto-generates argument schemas, but inspect complex types yourself.

Common errors: - "server disconnected": process died at startup. Check stderr. - "tool not found": registered but typo or stale cache after signature change. - "invalid arguments": client sent JSON that violates the schema. Use Inspector to inspect the exact argument shape.

5. Choosing a transport

Transport Use case Notes
stdio Local server, single user Simplest. Client spawns server as a child process
SSE / Streamable HTTP Remote server, multi-user HTTP-based with auth. For data centers / internal services

Start with stdio. Move to HTTP only when remote use becomes a real requirement.

6. Security checklist — Before trusting an MCP server

Third-party MCP servers carry arbitrary code-execution authority. A tool call is a function execution, and the function can do anything. Before connecting an external server:

  • Read the source. Open the GitHub repository and check what system calls the server makes.
  • Permission scope. File system access, network, external API keys — what does it require?
  • Isolation. Use containers, VMs, or separate OS users when possible.
  • Audit logging. Record every tool invocation.

Use catalogs like awesome-mcp-servers as a starting point, but run all four checks before any production connection.

7. At a glance

Step Core Watch out for
Design Separate Tools / Resources / Prompts Side-effecting → Tool, side-effect-free → Resource
Implementation mcp SDK + @mcp.tool() decorator Never print to stdout
Registration claude mcp add <name> --command ... Use absolute paths
Debugging MCP Inspector + stderr logs "server disconnected" → check stderr
Operation Permission scope + isolation + audit External MCP = arbitrary code execution

Next up

Part 4/5: Multi-Agent Patterns — Orchestrator and Specialist Separation. If MCP integrates external tools, multi-agent patterns are how you split internal work.

References

  • Model Context Protocol, Specification — modelcontextprotocol.io (verified 2026-05-05).
  • Anthropic, Using MCP with Claude Code — code.claude.com/docs/mcp (verified 2026-05-05).
  • awesome-mcp-servers — github.com/punkpeye/awesome-mcp-servers.

This is part 3/5 of the Coding Agents in Practice series.

๋Œ“๊ธ€

๋Œ“๊ธ€ ์“ฐ๊ธฐ

์ด ๋ธ”๋กœ๊ทธ์˜ ์ธ๊ธฐ ๊ฒŒ์‹œ๋ฌผ

Agent Memory Engine (2/10) — Building an AI Agent Memory System with SQLite Alone

← 1/10 SQLite Memory Engine f… ๐Ÿ“š Series Index 3/10 memcore Completed → A memory engine that runs without daemons, without dependencies, anywhere ํ•ต์‹ฌ ์š”์•ฝ This post covers how to design an AI agent memory engine (memcore) on a single-file SQLite backend. - Structure: 25 modules, ~6,464 lines, 22 DB tables - Principles: no daemon, single-file portability, no required external dependencies, host-agnostic - Core techniques: 4-layer hybrid search (topic · vector · FTS5 · LIKE), U-tag dialectic-based confidence evolution, ingestion-layer bias rejection rules What This Post Covers One approach to building AI agent memory without a server or cloud vector DB. Specifically: (1) the limitations of text-file-based memory, (2) vector DB alternatives compared and selection criteria, (3) how to layer search tiers, (4) how to evolve confidence scores over time, and (5) how to block biased memory accumulation at the ingestion stage. Design Principles: Why Single-File SQLite Text-file-based ...
Read more »

"ML Foundations (9/9) — PyTorch vs TensorFlow, and the Road to Local LLMs"

์ด๋ฏธ์ง€
← 8/9 Deep Learning Architec… ๐Ÿ“š Series Index (series end) The final part. So far we've focused on models. Now we focus on the tools that actually run them . We'll lay out the philosophical difference between PyTorch and TensorFlow, and trace how Hugging Face transformers , llama.cpp , MLX , and Ollama built the bridge to running large language models on your own machine. By the end you should have the full mental model of "download a pretrained LLM and serve it locally." 0. Learning Objectives Explain the eager-vs-graph difference between PyTorch and TensorFlow. Explain, in graph terms, how autograd automates the backward pass. Use Hugging Face transformers ' AutoModel , AutoTokenizer , and pipeline abstractions. Describe llama.cpp's GGUF quantization, INT4 inference, and CPU-first flow. Describe Apple MLX's use of unified memory and how it differs from PyTorch. Run a local LLM with Ollama and call it through an OpenAI-compatible API. ...
Read more »

"RAG Core Study (14/26) — Evaluation Sets with RAGAS & DeepEval"

์ด๋ฏธ์ง€
Series overview: Series index ๐Ÿ“š Series Index Without an evaluation set, every RAG improvement is a story, not evidence. RAG systems fail in more than one place: retrieval, context selection, answer generation, and grounding. That means "looks better" is not a metric. Part 14 explains how to build a golden dataset , how RAGAS and DeepEval fit into the workflow, and why teams must separate retrieval quality from answer quality if they want tuning decisions to hold up. 0. Prerequisites Part 13 reranking — system quality now depends on multi-stage ranking. Part 1 grounding — a correct-looking answer is not enough. Part 12 Hybrid — multiple retrieval settings need comparison on fixed data. 1. Learning Objectives Build a small but useful golden dataset for RAG. Distinguish retrieval metrics from answer metrics . Use RAGAS and DeepEval in the right roles. Avoid the common traps in judge-model-based evaluation. 2. ํ•ต์‹ฌ ์š”์•ฝ An evaluation set for RAG is ...
Read more »

"ML Foundations (8/9) — Deep Learning Architectures: CNN, RNN, Attention"

์ด๋ฏธ์ง€
← 7/9 Deep Learning Training ๐Ÿ“š Series Index 9/9 PyTorch vs TensorFlow → What did we keep adding on top of an MLP? Vision went CNN, sequences went RNN/LSTM, and eventually both converged on Attention and the Transformer. Each architecture is easier to remember if you read it as what it gave up to gain something else . This part is the one-line summary of the decisive inflection points: LeNet → AlexNet → ResNet → LSTM → Attention → Transformer. 0. Learning Objectives Explain why CNN's convolution, pooling, and weight sharing match images so well. Trace what got unblocked from LeNet → AlexNet → VGG → ResNet as depth grew. State BPTT for RNNs and the vanishing/exploding gradient problem. Write the LSTM gate equations (forget, input, output) with intuition. Write the attention formula in query/key/value form and the scaled dot-product variant. State the precise reasons Transformers replaced RNNs (parallelism + long-range dependencies). 1. ํ•ต์‹ฌ ์š”์•ฝ CNN : weight sharin...
Read more »

"ML Foundations (7/9) — Deep Learning Training: Optimizers, Regularization, Initialization"

← 6/9 Neural Networks ๐Ÿ“š Series Index 8/9 Deep Learning Architec… → Building an MLP, as we did in Part 6, is not the same as getting it to train well . The right optimizer, regularization, initialization, and learning rate — when these line up, deep networks converge. When they don't, the network refuses to learn at all. This part is the catalog of those crafts, with formulas, paper citations, and code in one place. 0. Learning Objectives Compare and write the update rules for SGD, Momentum, Nesterov, and Adam. Explain how Dropout, BatchNorm, and LayerNorm work and where they belong in a model. Derive the variance formulas for Xavier and He initialization and match them to activations. Implement step, cosine, and warmup learning-rate schedules in PyTorch. Explain why gradient clipping is effectively required for RNNs and Transformers. Diagnose the most common training failures (NaN, plateau, overfitting) and apply first-line fixes. 1. ํ•ต์‹ฌ ์š”์•ฝ SGD : \(w \leftarro...
Read more »

OpenClaw to Hermes Migration (2/13) — What to Preserve, Partially Port, or Discard

← 1/13 Current Structure Inve… ๐Ÿ“š Series Index 3/13 What Moves to Hermes → A code review record classifying operational scripts into three tiers: preserve, partial-preserve, and discard. What This Post Covers An asset classification framework (preserve / partial-preserve / discard) for migrating a legacy agent system to a successor A concrete case applying the Strangler Fig + Subset Migration strategy to a memory pipeline Classification results evaluated across LOC, external dependencies, and replaceability axes A method for slimming down a bloated single file (1,310 LOC) using the migration as a natural entry point Problem Definition: Port Everything or Cut Deep? The OpenClaw inventory comprises multiple agents, multiple scheduler/daemon processes, and multiple operational scripts. When moving to the successor system, Hermes, the choice is not binary. A full port carries accumulated technical debt along with it; a minimal port discards hard-won domain logic. Code-revie...
Read more »

AI Agents I Built (5/7) — Building an Automated Blogger API Publishing System

← 4/7 My Life Organizer Agent ๐Ÿ“š Series Index 6/7 Dual → Designing a Markdown-Based Auto-Publish Pipeline with Blogger API v3 + OAuth 2.0 Key Summary Blogger API v3 + OAuth 2.0 enables a pipeline that converts Markdown files into publishable HTML posts blogger_publish.py handles the full flow — frontmatter parsing → body cleanup → HTML conversion → CSS injection → API post — in a single script For bulk publishing, quota management (delay, retry, TOC suppression) is the critical design concern Platform Selection Criteria API access is mandatory for AI agent-driven content publishing. A comparison of platforms: Platform API Cost AdSense Decision WordPress.com REST API Paid plan required Paid only Cost overhead Ghost REST API Self-hosted or paid Manual setup Ops overhead Blogger v3 API Free Native integration Adopted Blogger is free, exposes a REST API, and integrates natively with AdSense. For personal blog automation, this combination i...
Read more »