Skip to main content
Glama
deenesjakoruzh
kritikatripathi03

toy-mcp-server

by kritikatripathi03
OverviewSchemaSourceRelated ServersDiscussionsAdmin
TypeScript
Local

toy-mcp-server

Минимальный сервер Model Context Protocol (MCP), написанный с нуля, чтобы понять, как MCP работает на самом деле, — без генераторов шаблонного кода, без шаблонов, только SDK.

Это первый практический MCP-проект: два инструмента и один ресурс, подключённые к Claude Desktop через stdio.

Что он делает

MCP позволяет LLM-клиенту (например, Claude Desktop) вызывать функции, работающие на вашем компьютере, а не просто генерировать текст на основе обучающих данных. Этот сервер предоставляет:

Инструменты (функции, которые может вызывать Claude):

  • roll_dice — бросает N кубиков с настраиваемым числом граней

  • flip_coin — подбрасывает монету N раз

Ресурсы (данные только для чтения, которые Claude может получать):

  • server-info — базовые метаданные о сервере (имя, назначение, время запуска)

Related MCP server: Random-Generator

Зачем это существует

LLM не имеют доступа ни к чему за пределами собственных обучающих данных и по умолчанию не могут ничего делать — они лишь генерируют текст. MCP — это стандартный способ дать модели:

  • возможность, которой у неё нет самой по себе (например, настоящую случайность — LLM, как известно, сами плохо справляются с выбором случайных чисел)

  • доступ к актуальным или приватным данным, о которых модель никак не может знать

Это маленькая безопасная песочница для этой идеи, прежде чем направлять MCP-сервер на что-то реальное (базу данных, API с авторизацией и т.д.).

Технологический стек

  • TypeScript

  • @modelcontextprotocol/sdk — официальный MCP SDK

  • zod — проверка схемы аргументов инструментов во время выполнения

  • transport stdio (локальное взаимодействие с Claude Desktop через дочерний процесс)

Структура проекта

toy-mcp-server/
├── src/
│   └── index.ts       # server setup, tools, and resource
├── build/              # compiled output (git-ignored)
├── package.json
├── tsconfig.json
└── README.md

Установка

1. Установите зависимости

npm install

2. Сборка

npm run build

3. Подключение к Claude Desktop

Найдите файл конфигурации Claude Desktop:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

  • Windows: обычно в %LOCALAPPDATA%\Packages\<Claude package folder>\LocalCache\Roaming\Claude\claude_desktop_config.json — точный путь зависит от способа установки (Microsoft Store или прямой установщик). Надёжный способ найти файл: откройте Claude Desktop → Настройки → Разработчик → Локальные MCP-серверы → Изменить конфигурацию — откроется именно тот файл, который читает приложение.

Добавьте этот сервер в раздел mcpServers (объедините с существующим файлом, а не перезаписывайте его):

{
  "mcpServers": {
    "toy-mcp-server": {
      "command": "node",
      "args": ["/absolute/path/to/toy-mcp-server/build/index.js"]
    }
  }
}

Замените путь на реальный абсолютный путь к build/index.js на вашей машине. В Windows экранируйте обратные слэши (\\) в JSON-строке.

4. Перезапустите Claude Desktop

Полностью завершите работу приложения (не просто закройте окно) и откройте снова. Проверьте Настройки → Developer → Локальные MCP-серверы, чтобы убедиться, что toy-mcp-server отображается как подключённый.

5. Попробуйте

В чате спросите:

  • "Брось 3 шестигранных кубика"

  • "Подбрось монету 10 раз"

Над ответом вы увидите небольшой индикатор вызова инструмента (например, «Бросить Dice»), подтверждающий, что Claude действительно выполнил функцию, а не угадал ответ.

Как это работает, кратко

  • McpServer — объект, который объявляет возможности сервера любому подключающемуся клиенту

  • registerTool(name, config, handler) — регистрирует вызываемую функцию. config.inputSchema использует Zod для проверки любых аргументов, отправленных моделью, прежде чем выполнится обработчик

  • registerResource(name, uri, config, handler) — регистрирует данные только для чтения, доступные по URI и получаемые без аргументов

  • StdioServerTransport — формат обмена: Claude Desktop запускает этот файл как до ашний процесс и общается через stdin/stdout по JSON-RPC. Здесь console.log никогда не используется для логирования, потому что stdout — это настоящий канал протокола; вместо него используется console.error (stderr)

Что дальше

  • Замените игрушечные инструменты на реальные, работающие с настоящей базой данных (Postgres через Prisma)

  • Добавьте инструмент на базе GitHub (например, list_open_prs), чтобы попрактиковаться в авторизации по токенам

  • Изучите transport Streamable HTTP, чтобы размещать этот сервер удалённо, а не локально

Лицензия

MIT

Available Tools

2 tools
flip_coinFlip CoinB

Flip a coin one or more times

ParametersJSON Schema
NameRequiredDescriptionDefault
timesNoNumber of flips

TDQS

B3.2/5.0
Behavior2/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations, the description carries the full burden of disclosing behavior. It merely says 'Flip a coin one or more times' without mentioning that the outcome is random, what the return format is (e.g., 'heads'/'tails'), or any side effects. The agent is left to guess the result structure.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is a single, efficient sentence that states the action and the parameter scope. There is zero wasted text, and the core information is front-loaded. It is appropriately concise for a trivial tool.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness2/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given there is no output schema, the description should explain what the tool returns (e.g., results of each flip). It omits any mention of return values or randomness, leaving the agent with incomplete information to use the tool correctly. For a tool this simple, more context is still needed.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

The single parameter 'times' is fully described in the schema ('Number of flips'), covering 100% of parameters. The tool description repeats the concept ('one or more times') but adds no extra meaning beyond what the schema already provides, so the baseline score of 3 applies.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description states a specific verb and resource ('Flip a coin') and adds the scope 'one or more times', which distinguishes it from the sibling roll_dice by naming the object of the action. An agent can clearly identify what this tool does.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines2/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

No guidance is given about when to use this tool versus roll_dice. The description only states the action, leaving the agent to infer from the name that this is for coins rather than dice. No exclusions, prerequisites, or alternative-routing hints are provided.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

roll_diceRoll DiceB

Roll one or more dice with a given number of sides

ParametersJSON Schema
NameRequiredDescriptionDefault
countNoNumber of dice to roll
sidesNoNumber of sides per die

TDQS

B3.2/5.0
Behavior2/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations provided, the description carries full responsibility for behavioral disclosure. It does not mention that the result is random, that it is a safe non-mutating operation, or any other behavioral trait beyond the literal act of rolling. This is a significant gap for a tool whose entire purpose is random generation.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is a single concise sentence with no waste. The core action is front-loaded. It is appropriately sized for a tool with only two parameters, though it could have been slightly more informative without losing conciseness.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness3/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a simple tool with two parameters fully documented in the schema, the description is reasonably complete. However, the lack of any mention of randomness or return behavior (even though no output schema is provided) leaves a small but notable gap. An agent might not know the outcome is random without prior knowledge.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so the schema already documents both parameters (count and sides) with defaults, minimums, and maximums. The description adds no semantic meaning beyond what the schema provides, which meets the baseline of 3.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description states exactly what the tool does: 'Roll one or more dice with a given number of sides'. It uses a specific verb (roll) and resource (dice), and clearly distinguishes from the sibling flip_coin, since dice and coins are different objects.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines2/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

No guidance on when to use this tool versus the sibling flip_coin. The context implies dice vs coin, but the description does not explicitly state that this is for dice and flip_coin is for coins, nor does it mention any exclusions or alternative conditions.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

  1. 2 tool updatesv1.0.0
    • First observedflip_coin
    • First observedroll_dice
Deploy Server
One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

TDQS

A3.6/5.0

Scored across 2 tools

Disambiguation5/5

The two tools, roll_dice and flip_coin, are completely distinct in their purpose and behavior. There is no overlap or ambiguity, making it trivial for an agent to select the correct tool.

Naming Consistency5/5

Both tool names follow the same verb_noun pattern (roll_dice, flip_coin), using snake_case and clear action-object structure. This is perfectly consistent and predictable.

Tool Count4/5

With only 2 tools, the server is minimal, but given its explicit 'toy' purpose and narrow domain (random chance events), this count is appropriate and focused. It is slightly thin by general standards but well-scoped for a toy.

Completeness4/5

The tool surface covers the two most common random generators (dice and coin flips). While a generic random number generator or card draw might be expected in a broader utilities server, for a toy server focused on simple chance, these tools are sufficient and leave no obvious dead ends.

Maintenance

ActivitySlowing
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • Random Number MCP

    zazencodes
    A
    license
    A
    quality
    C
    maintenance
    Production-ready MCP server that provides LLMs with essential random generation abilities, including random integers, floats, choices, shuffling, and cryptographically secure tokens.
    7
    54 PyPI
    51
    MIT

  • Random-Generator

    xiaobenyang-com
    A
    license
    A
    quality
    D
    maintenance
    An encrypted and secure random number generation server that complies with the MCP protocol, suitable for AI applications, LLMS, and other systems that require high-quality random numbers.
    7
    2
    Apache 2.0

  • mcp-tools-server

    RenanMiqueloti
    A
    license
    Not graded
    quality
    D
    maintenance
    A general-purpose MCP server with utility tools including datetime information, safe math calculations, text statistics, JSON extraction, knowledge base search, and HTTP GET requests. It demonstrates server-side MCP implementation and can be connected to Claude Desktop or LangGraph agents.
    MIT

  • mcp-random

    MikeyBeez
    A
    license
    A
    quality
    C
    maintenance
    MCP server providing true randomness capabilities to Claude, enabling cryptographically secure random number generation for games, decision-making, sampling, simulations, and any operation requiring genuine randomness.
    13
    2
    MIT