Back to Skills

build-custom-mcp-server

pjt222
Updated 6 days ago
13 views
17
2
17
View on GitHub
Metaaitestingapimcpdesign

About

This skill enables developers to build custom MCP servers to expose domain-specific tools to AI assistants. It covers implementation in Node.js or R, tool definition, and transport configuration for integrating specialized functionality or wrapping existing APIs. Use it when you need capabilities beyond standard mcptools or want to create specialized AI integrations.

Quick Install

Claude Code

Recommended
Primary
npx skills add pjt222/agent-almanac -a claude-code
Plugin CommandAlternative
/plugin add https://github.com/pjt222/agent-almanac
Git CloneAlternative
git clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/build-custom-mcp-server

Copy and paste this command in Claude Code to install this skill

Documentation

建客 MCP 服

造客 MCP 服以露域專工於 AI 輔。

  • 當露客能於 Claude Code 或 Claude Desktop
  • 建 mcptools 之外專工
  • 造域專 AI 輔整
  • 包現 API 或服為 MCP 工

  • :當露工列(名、述、參、行)
  • :施語(Node.js 或 R)
  • :傳類(stdio 或 HTTP)
  • :認需
  • :Docker 包需

一:定工規

書碼前→每工定:

tools:
  - name: query_database
    description: Execute a read-only SQL query against the analysis database
    parameters:
      query:
        type: string
        description: SQL SELECT query to execute
        required: true
      limit:
        type: integer
        description: Maximum rows to return
        default: 100
    returns: JSON array of result rows

  - name: run_analysis
    description: Execute a predefined statistical analysis by name
    parameters:
      analysis_name:
        type: string
        description: Name of the analysis to run
        enum: [descriptive, regression, survival]
      dataset:
        type: string
        description: Dataset identifier
        required: true

得: 每工以 YAML 或 markdown 規附名、述、參(含類、默、必旗)、返類已錄於書碼前。

敗: 工規不清→訪域專家或察現 API 文以定參類與返格。

二:施於 Node.js(用 MCP SDK)

// server.js
import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import { z } from "zod";

const server = new McpServer({
  name: "my-analysis-server",
  version: "1.0.0",
});

// Define tools
server.tool(
  "query_database",
  "Execute a read-only SQL query against the analysis database",
  {
    query: z.string().describe("SQL SELECT query"),
    limit: z.number().default(100).describe("Max rows to return"),
  },
  async ({ query, limit }) => {
    // Validate read-only
    if (!/^\s*SELECT/i.test(query)) {
      return {
        content: [{ type: "text", text: "Error: Only SELECT queries allowed" }],
        isError: true,
      };
    }

    const results = await executeQuery(query, limit);
    return {
      content: [{ type: "text", text: JSON.stringify(results, null, 2) }],
    };
  }
);

server.tool(
  "run_analysis",
  "Execute a predefined statistical analysis",
  {
    analysis_name: z.enum(["descriptive", "regression", "survival"]),
    dataset: z.string().describe("Dataset identifier"),
  },
  async ({ analysis_name, dataset }) => {
    const result = await runAnalysis(analysis_name, dataset);
    return {
      content: [{ type: "text", text: JSON.stringify(result, null, 2) }],
    };
  }
);

// Start server with stdio transport
const transport = new StdioServerTransport();
await server.connect(transport);

得: 可行 server.js 檔—引 MCP SDK、以 Zod 綱定工、以 stdio 傳連。node server.js 無誤始服。

敗:@modelcontextprotocol/sdkzod 已裝(npm install)。察引徑合 SDK 版(SDK 於版間重排導出)。

三:施於 R(用 mcptools)

# server.R
library(mcptools)

# Register custom tools
mcp_tool(
  name = "query_database",
  description = "Execute a read-only SQL query",
  parameters = list(
    query = list(type = "string", description = "SQL SELECT query"),
    limit = list(type = "integer", description = "Max rows", default = 100)
  ),
  handler = function(query, limit = 100) {
    if (!grepl("^\\s*SELECT", query, ignore.case = TRUE)) {
      stop("Only SELECT queries allowed")
    }
    result <- DBI::dbGetQuery(con, paste(query, "LIMIT", limit))
    jsonlite::toJSON(result, auto_unbox = TRUE)
  }
)

# Start server
mcptools::mcp_server()

得: 可行 server.R 檔—以 mcp_tool() 註客工、以 mcp_server() 始服。Rscript server.R 始 MCP 服。

敗:mcptools 自 GitHub 裝(remotes::install_github("posit-dev/mcptools"))。察處函簽合參定。

四:設案構

my-mcp-server/
├── package.json          # Node.js dependencies
├── server.js             # Server implementation
├── tools/                # Tool implementations
│   ├── database.js
│   └── analysis.js
├── test/                 # Tests
│   └── tools.test.js
├── Dockerfile            # Container packaging
└── README.md             # Setup instructions

得: 案目造附 server.js(或 server.R)、package.jsontools/ 目以分模工、test/ 目以試。

敗: 目構不合施語→調。R 服或用 R/tools/tests/testthat/test/

五:試服

stdio 手試

echo '{"jsonrpc":"2.0","method":"tools/list","id":1}' | node server.js

註於 Claude Code

claude mcp add my-server stdio "node" "/path/to/server.js"

驗工現

起 Claude Code 會而察客工列且可行。

得: tools/list JSON-RPC 呼返諸定工附正名與綱。claude mcp list 示服已註。工於 Claude Code 會可呼。

敗: tools/list 返空陣→工未註於 server.connect() 前。Claude Code 不得服→驗 claude mcp add 命徑絕而二進可行。

六:加誤處

server.tool("risky_operation", "...", schema, async (params) => {
  try {
    const result = await performOperation(params);
    return {
      content: [{ type: "text", text: JSON.stringify(result) }],
    };
  } catch (error) {
    return {
      content: [{ type: "text", text: `Error: ${error.message}` }],
      isError: true,
    };
  }
});

得: 每工處以 try/catch 包。誤入返 isError: true 附述訊非崩服程。

敗: 服仍崩於劣入→察 try/catch 包全處體含諸 async 操。確保 promise 於 try 區內 await。

七:包發

package.json 附 bin 條:

{
  "name": "my-mcp-server",
  "version": "1.0.0",
  "bin": {
    "my-mcp-server": "./server.js"
  },
  "dependencies": {
    "@modelcontextprotocol/sdk": "^1.0.0",
    "zod": "^3.22.0"
  }
}

用者可裝與配:

npm install -g my-mcp-server
claude mcp add my-server stdio "my-mcp-server"

得: package.jsonbin 條指服入。用者以 npm install -g 全裝與 claude mcp add 註。

敗: bin 全裝後不行→確 server.js 首行有 shebang(#!/usr/bin/env node)且可行。驗包名不撞現 npm 包。

  • 服無誤始
  • tools/list 返諸定工附正綱
  • 每工以有效入正行
  • 工於誤入返宜誤
  • 服以 stdio 傳合 Claude Code
  • 工於 Claude 會可發現且可用

  • 阻操:MCP 服當異處請。久操阻他工呼。
  • 缺誤處:未處異崩服。必以 try/catch 包工處。
  • 綱不合:工參綱當全合處所候
  • stdio 緩:用 stdio 傳→確出已沖。Node.js 默緩 stdout。
  • :MCP 服與程同權。精驗入,尤於殼命或庫查。

  • configure-mcp-server — 連已建服與客
  • troubleshoot-mcp-connection — 除連題
  • containerize-mcp-server — 以 Docker 包服

GitHub Repository

pjt222/agent-almanac
Path: i18n/wenyan-ultra/skills/build-custom-mcp-server
0
agentsagentskillsai-assisted-developmentclaude-codeskillsteams

Related Skills

content-collections

Meta

This skill provides a production-tested setup for Content Collections, a TypeScript-first tool that transforms Markdown/MDX files into type-safe data collections with Zod validation. Use it when building blogs, documentation sites, or content-heavy Vite + React applications to ensure type safety and automatic content validation. It covers everything from Vite plugin configuration and MDX compilation to deployment optimization and schema validation.

View skill

polymarket

Meta

This skill enables developers to build applications with the Polymarket prediction markets platform, including API integration for trading and market data. It also provides real-time data streaming via WebSocket to monitor live trades and market activity. Use it for implementing trading strategies or creating tools that process live market updates.

View skill

creating-opencode-plugins

Meta

This skill helps developers create OpenCode plugins that hook into 25+ event types like commands, files, and LSP operations. It provides the plugin structure, event API specifications, and implementation patterns for JavaScript/TypeScript modules. Use it when you need to intercept, monitor, or extend the OpenCode AI assistant's lifecycle with custom event-driven logic.

View skill

sglang

Meta

SGLang is a high-performance LLM serving framework that specializes in fast, structured generation for JSON, regex, and agentic workflows using its RadixAttention prefix caching. It delivers significantly faster inference, especially for tasks with repeated prefixes, making it ideal for complex, structured outputs and multi-turn conversations. Choose SGLang over alternatives like vLLM when you need constrained decoding or are building applications with extensive prefix sharing.

View skill