This MCP server provides three tools for Moonlight Garden: run a daily routine, get a read-only farm briefing, and check cookable recipes.
daily_routine: manually run 3 garden jobs, buy 10 bread worms, cast 10 times, and optionally fetch the latest top 5 recipes plus update a local expensive-recipe ledger. Uses UTC idempotency; unknown outcomes require resolveUnknown (assume_completed or retry) with explicit user choice.
farm_brief: read-only summary of farm plots, crop maturation, watering cooldown, chicken coop/eggs, cabbage gap, beehives, bee trips, and waiting young bees. Performs no garden or bee actions; returns UTC timestamps.
recipe_check: read local ledger, live inventory, and remaining cook quota; rank cookable recipes by value added, list missing-one recipes, and fall back to the latest 50 recipes if the ledger has no cookable entries. Never cooks or sells; may auto-claim pending gifts via inventory list and return UTC receipts.
Configure enabled tools, recipe ledger storage, and cabbage-gap formula; connect via MOONLIGHT_GARDEN_MCP_URL and optional MOONLIGHT_GARDEN_MCP_TOKEN. No timers or background execution.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@moonlight-garden-handytools run my daily garden routine"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
月光花园游戏地址：https://moonlightgarden.space/
一组独立、非官方的月光花园 MCP 小工具。它通过用户自己的月光花园 MCP 连接工作，不负责创建账号或提供连接凭据。
本项目仅用于减少 Agent 执行花园日常时的工具调用次数，并聚合、精简返回数据；工具不会自行运行，也不代替 Agent 或用户作出游戏决策。
三个工具可以分别启用：
daily_routine：把打零工 3 次、购买 10 个小面包虫、钓鱼 10 次合并为一次手动调用；菜谱功能开启时，还会读取最新 50 道菜、选出售价最高的 5 道并更新本地贵菜账。它保留每一步的原始结果，并用 UTC 日期和本地状态避免同一天误调用时重复执行。
farm_brief：只读汇总可耕地与作物成熟时间、浇水冷却、鸡舍与鸡蛋、白菜缺口，以及蜂箱、采蜜出行和门口幼蜂状态；不会执行任何花园或蜜蜂动作。
recipe_check：读取本地贵菜账、实时背包和今日剩余做菜次数，找出材料齐全或只缺一种材料的菜，并按单次增值排序；贵菜账没有可做菜时才检查最新 50 道作为兜底。它只验证和排序，不会自动做菜或卖菜。
这套工具内置了一条简单的菜谱工作流，适合希望更快积累 Moon 的玩家（快速积累 Moon 会加速消耗游戏内容，请用户与 Agent 充分讨论后再决定是否保留）：
每天手动调用 daily_routine 后，工具会自动读取最新 50 道菜谱，从中选出售价最高的 5 道，并按上游提供的权威 dishItemId 去重更新到本地贵菜账。
做菜前调用 recipe_check，工具会把贵菜账与实时背包、今日剩余做菜次数进行匹配，列出当前材料齐全的菜，并按单次增值排序。
尽量优先制作 recipe_check 排在前面的高增值菜，有助于更快积累 Moon。这里比较的是成品售价减去原料直接出售价值后的增值，不是只看成品售价。
daily_routine没有定时器，也不会在每天某个时间自动运行；只有用户或 Agent 主动调用时，上述菜谱读取和记录才会执行。
如果不需要这套菜谱方案，请把 recipeLedger.enabledInDailyRoutine 设为 false，并从 enabledTools 中移除 recipe_check。这样 daily_routine 仍可继续执行打零工、买虫和钓鱼，farm_brief 也不受影响。
Related MCP server: hayday-mcp-server
以下几项都可以直接在配置文件中修改：
需求 | 配置项 | 作用 |
开关菜谱拉取与贵菜账记录 |
| 一个总开关；开启时两项一起运行，关闭时两项一起停用 |
修改贵菜账保存目录 |
| 可填写本机绝对路径；设为 |
修改贵菜账文件名 |
| 可自行命名，但只能填写文件名，不能在这里夹带目录路径 |
修改白菜缺口公式 |
| 可用鸡数量和背包白菜数量组合出适合自己的计算规则 |
recipe_check 依赖本地贵菜账。若账本不存在或格式错误，它会明确返回账本读取错误，不会假装“当前没有能做的菜”。
Node.js 20 或更新版本
可用的月光花园 Streamable HTTP MCP 地址
git clone https://github.com/shikihuang04/moonlight-garden-handytools.git
cd moonlight-garden-handytools
npm ci
npm run build
cp config.example.json moonlight-garden-handytools.config.json连接信息只从环境变量读取：
MOONLIGHT_GARDEN_MCP_URL：必填，月光花园 MCP 地址。公网地址必须使用 HTTPS。
MOONLIGHT_GARDEN_MCP_TOKEN：可选。如果你的连接使用 Bearer token，在这里提供。
项目不会把连接地址或 token 写进贵菜账、幂等记录或日志。
以下是通用 stdio 配置形状；把路径替换为你的绝对路径：
{
"mcpServers": {
"moonlight-garden-handytools": {
"command": "node",
"args": [
"/absolute/path/moonlight-garden-handytools/dist/index.js",
"--config",
"/absolute/path/moonlight-garden-handytools/moonlight-garden-handytools.config.json"
],
"env": {
"MOONLIGHT_GARDEN_MCP_URL": "https://your-moonlight-garden-mcp.example/mcp",
"MOONLIGHT_GARDEN_MCP_TOKEN": "your-token-if-required"
}
}
}
}不同客户端的外层配置键可能不同，但启动命令、参数和环境变量相同。
如果你的连接不需要 Bearer token，请把整个 MOONLIGHT_GARDEN_MCP_TOKEN 条目删除，不要填写示例文字。保存后重载或重启 MCP 客户端。
这是绕过工具面板的另一种调用方式，不是修复 Operit 管理页。 Operit 1.12.2（versionCode 51）的反馈来自 Android / proot Ubuntu、Node.js v20.20.2：管理页仍“验证失败”，工具没有进入 Agent 的可用列表；反馈者通过自己的终端脚本成功调用过 farm_brief，但没有验证另外两个工具。
下面使用本仓库提供的 scripts/call-tool.mjs，不需要反馈者的私人脚本。它完成 MCP 连接、调用一次指定工具、输出结果并关闭连接，不自动重试，也不会注册到 Operit 的工具面板。此示例已在本地 macOS / Node.js 24 上用真实 stdio 服务和模拟花园验证；尚未在 Android / Operit 上实测本示例。已能正常接入或已有可用绕行脚本的用户，无需更换。
使用 Linux 终端中已完成 npm ci、npm run build 的项目目录，其中应有 dist/index.js、node_modules/ 和 scripts/call-tool.mjs。第一次安装按上方“要求与安装”操作。旧版已构建成功的用户，只需在同一项目添加本仓库的 scripts/call-tool.mjs；不要覆盖现有配置或运行数据。
以下命令均在该项目目录执行，不在 Android 下载目录的另一份副本操作。不需要重装 Operit、修改 mcp_config.json、移动账本或修改其他插件。
在项目根目录用本地编辑器创建 .env，填入自己的连接信息（不要把真实凭据贴进聊天或终端命令历史）：
MOONLIGHT_GARDEN_MCP_URL="https://your-moonlight-garden-mcp.example/mcp"
# 不需要 Bearer token 时省略下一行。
MOONLIGHT_GARDEN_MCP_TOKEN="your-token-if-required"chmod 600 .env.env 只保留在本机，已加入本仓库忽略规则；不要分享或提交。下方 --env-file 写法需要 Node.js 20.6 或更新版本。如果终端进程环境里已经正确设置了这两个变量，可以省略 --env-file=.env；已有进程环境变量优先于 .env，切换账号时要留意旧值。
下方 --config 必须指向你原本使用的配置文件，继续使用相同的 profileId 和数据目录，才能共用贵菜账与每日幂等记录。没有自建配置、且一直使用默认配置的用户可以省略 --config 及其值；不要为了绕行重新创建一个 profile。
只列出当前配置启用的工具，不调用上游花园：
node --env-file=.env scripts/call-tool.mjs list --config moonlight-garden-handytools.config.json需要查看农场时，单次调用只读 farm_brief：
node --env-file=.env scripts/call-tool.mjs farm_brief --config moonlight-garden-handytools.config.jsonlist 成功只证明本地连接与工具发现成功；farm_brief 返回实际农场状态且没有对应来源的 errors，才证明那些来源查询成功。已经有实测结果的用户不必为反馈重复调用。
若让 Agent 代为执行，需要它本身有终端执行能力及相应授权。可告诉它：“在我的 Linux 插件目录运行上面的 farm_brief 命令，读取 JSON 结果；不要调用 daily_routine。”这仍是终端命令，不是 Operit 工具列表中的直接调用，不会自动获得额外权限。
检查菜谱（依赖现有本地贵菜账；背包查询可能自动领取待领礼物，详见下方 recipe_check 说明）：
node --env-file=.env scripts/call-tool.mjs recipe_check --config moonlight-garden-handytools.config.json以下命令会真实打工、买鱼饵、钓鱼，并可能更新贵菜账。不要把它当接入测试。 仅在用户或 Agent 确实要执行整套日常时使用：
node --env-file=.env scripts/call-tool.mjs daily_routine --config moonlight-garden-handytools.config.json它仍复用原工具的启用开关、UTC 幂等保护和失败处理，不是另一套日常逻辑。需要处理 needs_resolution 时，可用 --args 传入 JSON 对象；例如 --args '{"resolveUnknown":"assume_completed"}'。只能在用户已理解风险并明确选择该恢复方式后添加，不能由 Agent 为了让命令成功而自行选择，详见“网络结果不确定时”。
标准输出是完整 MCP 结果 JSON：structuredContent 是工具数据，content 保留原始文本结果；失败时仍保留工具已经返回的成功部分。
isError: true 时退出码为 1；参数、启动或连接失败也返回非零退出码，错误写到标准错误。不要只看有没有 JSON，还要看 errors、status 和 failedStep。
调用最多等待 5 分钟，不自动重试。超时、断线或手动中断不代表游戏动作没有发生；不要盲目重复日常或删除幂等文件，应按原工具的未知结果恢复规则处理。
MOONLIGHT_GARDEN_MCP_URL is required. 表示启动进程缺少必填地址，本身不能证明 Operit 有环境变量传递故障。检查当前环境或 .env，不需要公开凭据。
run.sh 如果仅执行 node dist/index.js，只是启动服务，不会替你完成单次工具调用；不要将它与本节的调用示例混淆，也不要执行 node dist/index.js farm_brief 来代替调用示例。
停止使用绕行方式时，不再执行这些命令即可；没有额外后台服务或定时任务需要卸载，正常 MCP 接入配置也未被修改。
{
"profileId": "default",
"enabledTools": ["daily_routine", "farm_brief", "recipe_check"],
"recipeLedger": {
"enabledInDailyRoutine": true,
"directory": null,
"fileName": "moonlight-garden-expensive-recipes.md"
},
"farmBrief": {
"cabbagePlantGapFormula": "chickenCount * 2 - cabbageInInventory"
}
}enabledTools 决定 MCP 对外注册哪些工具。可以只保留一个或两个；不需要三个一起接入。
profileId 是用户自行填写的本地账号标签，不是月光花园真实用户 ID，也不会从 token 自动推导。允许英文字母、数字、点、下划线和连字符。每个花园账号必须使用不同的 profileId；更换连接凭据时也要同步检查它，否则不同账号会误用同一份每日状态与贵菜账。
recipeLedger.enabledInDailyRoutine 是一个总开关：
true：daily_routine 同时执行“读取最新 50 道并取售价前 5”和“更新贵菜账”。
false：两步一起关闭；打零工、买虫和钓鱼照常工作。
这个开关不禁用 recipe_check。recipe_check 本身依赖贵菜账；账本不存在或格式错误时会返回明确的 errors.ledger，不会拿空列表冒充“没有能做的菜”。
recipeLedger.directory 是所有 profile 的存储根目录。自定义值必须是绝对路径；不会展开 ~，相对路径会被拒绝。值为 null 时使用平台默认根目录：
macOS：~/Library/Application Support/moonlight-garden-handytools
Windows：%LOCALAPPDATA%/moonlight-garden-handytools
Linux：$XDG_DATA_HOME/moonlight-garden-handytools，未设置时为 ~/.local/share/moonlight-garden-handytools
实际运行文件放在 <存储根目录>/profiles/<profileId>/：
profiles/default/
├── daily-routine-state.json
├── daily-routine-state.json.lock
├── moonlight-garden-expensive-recipes.md
└── moonlight-garden-expensive-recipes.md.bak锁文件只在 daily_routine 执行期间存在，账本备份只在覆盖已有账本后出现。目录和账本文件名都可以在配置中修改；文件名必须是普通文件名，不能包含目录路径。
cabbagePlantGapFormula 可使用数字、chickenCount、cabbageInInventory、括号和 + - * /。它由受限解析器计算，不执行 JavaScript。默认公式为：
chickenCount * 2 - cabbageInInventory所有绝对时间均为 UTC ISO 8601，例如 2026-09-11T04:05:06.789Z。剩余秒数仍是相对时长，不受时区影响。需要本地提醒时，由调用方把 UTC 转成用户所在时区。
daily_routine 的幂等日期也按 UTC 00:00 切换。这是本工具自己的防误调用边界，不声称等同于月光花园服务器的地区自然日。
本项目没有定时器、heartbeat 或后台自动执行入口。只有 MCP 客户端调用 daily_routine 时，它才会运行；是谁或什么流程促使 Agent 发起调用，不属于工具限制。
构建并保存 MCP 配置后：
重载 MCP 客户端，确认工具列表中只出现 enabledTools 选择的工具。
如果启用了 farm_brief，第一次可先调用它，例如对 Agent 说：“调用 farm_brief 查看当前农场摘要，不执行任何花园动作。”成功结果应含以 Z 结尾的 UTC checkedAt；plots 不含房屋地块。某个来源失败时应出现对应 errors，而不是伪造空状态。
如果没有启用 farm_brief，确认工具列表正确即可；不必为了验收而临时启用或调用其他工具。
recipe_check 时初始化账本仓库内的 recipe-ledger.example.md 是空白账本模板。先在配置对应的 <存储根目录>/profiles/<profileId>/ 创建目录，再把模板复制为 recipeLedger.fileName 指定的文件名。
账本最小内容如下：
# Moonlight Garden expensive recipes
```json
[]
```空账本是有效账本：recipe_check 会因账本内没有可做候选而查询最新 50 道菜，并以 usedLatest50Fallback: true 返回兜底候选。但 recipe_check 永远不会写入贵菜账，因此单独使用它不会逐日积累菜谱；若要积累，请启用 daily_routine 的菜谱查询与写账功能，或自行维护该文件。
不要仅仅为了创建账本而调用 daily_routine：它会先真实执行打零工、买 10 个小面包虫和钓鱼 10 次。只有确实想执行整套日常时，才让它顺便建立或更新账本。
daily_routine按顺序执行：
garden_work 三次
fish({ command: "buy bread 10" })
fish({ command: "cast 10" })
总开关开启时读取 kitchen({ command: "recipes 50" })，按 sell_value 取前 5
总开关开启时按 dishItemId 更新本地贵菜账
每一步保留上游原始 result；菜谱步骤另带 topRecipes。贵菜账只保存：
dishItemId（上游 item，也是去重主键）
name
ingredientNames（上游中文材料名原样保留）
sellValue
账本按 sellValue 降序，写前备份为 .bak，再原子替换。
同一 UTC 日期再次调用时，已成功步骤不会重做。明确的 work_limit 或 casts_remaining=0 会记为 completed_by_daily_limit 并继续后续步骤。上游明确返回的普通错误会返回：
status: "partial_failure"
failedStep
此前成功步骤及其原始结果
未执行步骤的 status: "not_run"
打工、购买或钓鱼发出后，如果没有收到权威回包，工具无法判断动作究竟有没有成功。该步骤会保存为 outcome_unknown，总状态为 needs_resolution；当天后续调用不会自动重试，也不会继续剩余步骤。
Agent 必须先向用户说明两种选择各自的风险，再由用户明确选择一种恢复方式。不能从“允许每天运行”等长期授权推断选择，也不能由 Agent 自行代选：
{ "resolveUnknown": "assume_completed" }将该步骤记为 completed_by_user，不重复动作，然后继续后续步骤；如果原动作其实没有完成，这会跳过它。
{ "resolveUnknown": "retry" }明确重试该步骤。原动作可能已经成功，因此这个选项可能造成重复打工、重复购买或重复钓鱼，只有用户接受风险时才能使用。没有 needs_resolution 时传入这个参数会报错且不执行日常。
若已取到前 5 但账本写入失败，update_recipe_ledger 会单独失败；下次调用只重试写账，不重复之前的花园动作或菜谱查询。
如果某个花园动作已经成功、但紧接着本地幂等状态写入失败，工具会停止并以 failedStep: "idempotency_state" 返回内存中已发生的结果。写入前已保存的预执行状态仍会阻止自动重试；修复目录权限或磁盘问题后，下一次调用会要求用户按 outcome_unknown 明确处理。
farm_brief只调用 farm({ command: "status" })、coop({ command: "status" }) 和 bee({ command: "status" })，不执行种植、收获、浇水、喂鸡，也不执行 send / settle / keep / release / buy / expand 或替用户挑选幼蜂。
输出包括：
UTC checkedAt
仅可耕地的 plots，排除房屋地块
累计口径的 maturingWithinHours（1/2/4 小时）
waterCooldownSeconds、UTC waterAvailableAt
maturesIfWateredPlotIds：仅当前剩余 1 到 1800 秒的地块
鸡舍状态、白菜库存和配置公式算出的 cabbagePlantGap
独立的 bees：蜂箱数量与容量、今日剩余采蜜次数、出行三态、图鉴进度、蜂箱成员和门口待选幼蜂
maturingWithinHours 中保存的是地块编号，并采用累计口径：1 小时内成熟的地块也会出现在 2 小时和 4 小时列表。cabbagePlantGap 保留公式的原始正负值；默认公式下，正数表示缺少的白菜数量，0 表示刚好，负数表示背包白菜多于当前目标。
蜜蜂 tripState 只有 home / away / ready_to_settle。away 时若 status 提供权威 Unix back_at，工具会转换为 UTC backAt；上游只提供取整后的中文倒计时时，backAt 为 null，并通过 backInText 原样返回倒计时。home 和 ready_to_settle 的 backAt/backInText 均为 null。固定四小时的单次采蜜时长不进入简报，避免被误读为剩余时间。
hive 返回 beeId/index/name/label/colorId/patternId/preferenceId/preferredCrops；未命名的蜂保留 name:null。waitingYoung 只返回 status 能确认的 beeId/label/colorId/patternId/preferenceId/preferredCrops。父母信息和 newToCodex 只存在于 settle 回包，不会由本工具猜测或补造。
任一来源失败时，工具在 errors 中写明来源，并省略依赖该来源的字段，不用空数组、null 或 0 冒充查询成功。蜜蜂来源失败或出现无法识别的出行状态组合时，返回 errors.bee 并整个省略 bees，成功取得的农田与鸡舍字段仍保留。
recipe_check流程：
读取本地贵菜账。
读取厨房剩余额度和实时 inventory list。
用背包、农场目录和鱼类收藏中可核对的“名称 ↔ item ID”做材料映射。
有贵菜可做时只处理账本；账本没有任何 cookableNow 时才读取最新 50 道兜底，并设 usedLatest50Fallback: true。
cookableNow 和 missingOne 各最多返回前 5，按单次增值降序。未知、重名或重复材料不猜 ID，只计入聚合摘要：
{
"unresolvedRecipeCount": 31,
"unresolvedIngredientSummary": [
{ "name": "石涧雉", "recipeCount": 8 }
]
}增值公式：
ingredientSellValue = 每种配方材料一份的权威 sell_value 之和
valueAdded = sellValue - ingredientSellValue背包物品使用 inventory list 的实时 sell_value。农场目录只能补作物的 sell_value；种子购买价 price 永远不参与计算。缺失材料没有权威售价时返回 ingredientSellValue: null、valueAdded: null 和 valueAddedStatus: "missingIngredientSellValueUnavailable"。
canCookAnyNow 只有在厨房额度大于 0 且 cookableNow 非空时才为 true。工具不自动做菜、卖菜，也不建议种植、钓鱼或打猎方案。
cookableNow 只表示实时背包中的材料齐全，不单独代表今天还允许下锅；实际是否现在能做必须看 canCookAnyNow 和 remainingCookCount。
月光花园现有 inventory list 可能自动领取待领礼物，所以 recipe_check 严格来说不是完全只读。如果上游本次返回了 received_gifts，结果会额外包含：
{
"receivedGifts": [
{
"fromName": "小机",
"receivedAt": "2026-09-11T04:05:06.789Z",
"itemId": "fish_maple_fish",
"itemName": "枫月鱼",
"quantity": 1
}
]
}receivedAt 是本工具通过 inventory list 领取/观测到礼物的 UTC 时间，不是对方发送礼物的时间；上游当前没有提供权威发送时间。没有领取礼物时省略 receivedGifts。
npm test
npm run check
npm run build测试使用本地模拟上游，不会执行真实打工、购买、钓鱼、领奖或其他花园动作。
MIT。这是独立实现的非官方项目，与月光花园及其运营方无隶属关系。
Manually run three garden jobs, buy 10 bread worms, cast 10 times, and optionally fetch the latest recipe top five plus update the local ledger. Unknown action outcomes stop safely. The Agent must not choose a recovery option without the user's explicit decision.
| Name | Required | Description | Default |
|---|---|---|---|
| resolveUnknown | No | Only use when a prior result reports needs_resolution. Ask the user for an explicit choice after explaining that assume_completed may skip an action that did not finish, while retry may duplicate an action that did finish. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The annotations declare idempotentHint=true, but the description states that 'retry may duplicate an action that did finish' — exactly the additional environmental effect that idempotency rules out. This directly contradicts the declared annotation, so under the rubric the disclosure is scored 1 despite the otherwise useful 'unknown outcomes stop safely' guidance.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences, front-loaded with the sequence of actions, which is appropriate for a composite tool. However, the first sentence is a long comma-spliced run-on that packs five distinct actions plus an optional branch into one breath, which hurts scannability.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a multi-step routine with one optional parameter, the description covers the actions and the failure/recovery posture, which is the essential behavior. It gives no indication of what the routine returns or how partial results are reported, and there is no output schema to fill that gap.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100% and the resolveUnknown property already documents the assume_completed/retry trade-off and the need to ask the user, so the baseline is 3. The description reinforces when the parameter applies ('Unknown action outcomes stop safely') but adds little syntax or value beyond the schema text.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description names concrete actions (run three garden jobs, buy 10 bread worms, cast 10 times, optionally fetch the recipe top five and update the ledger), so the agent knows this is a composite routine rather than a single operation. It does not, however, differentiate itself from the siblings farm_brief or recipe_check, and 'three garden jobs' leaves the specific jobs unnamed.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The word 'Manually' implies this is the human-triggered counterpart to some automatic process, and the closing sentence gives a clear directive about when a recovery option may be used. There is no explicit comparison to farm_brief or recipe_check, so the agent must infer which sibling to prefer.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Read and compress farm, watering, chicken, cabbage-gap, bee-hive, bee-trip, and waiting-young state. It performs no garden or bee actions and returns UTC timestamps.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnly, idempotent, non-destructive. The description adds useful negative scope ('performs no garden or bee actions') and the return detail 'returns UTC timestamps', which the annotations do not cover. Without an output schema, more on the compressed shape would help, so this is a moderate add.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences, zero waste, front-loaded with the verb and fully enumerated scope. The negative-scope clause is placed after the positive read statement, which is the right order.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
A parameterless read tool with strong annotations and no output schema: the description covers what is read, what is not done, and one return characteristic (UTC timestamps). Slightly more on the compressed output format would make it complete.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Zero parameters, so baseline 4. The description correctly implies the call is parameterless by describing an unconditional state read, with no misleading param claims.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb+resource: 'Read and compress' the farm state across named domains (farm, watering, chicken, cabbage-gap, bee-hive, bee-trip, waiting-young). The scope is concrete enough that an agent knows what it returns, though siblings daily_routine and recipe_check are not referenced for differentiation.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No explicit when-to-use guidance and no mention of the sibling tools daily_routine or recipe_check. The agent must infer that this is a broad state-summary call from the domain list alone.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Validate and rank cookable ledger recipes using live inventory and quota. It never cooks or sells, but the upstream inventory list may auto-claim pending gifts; claimed gifts are returned with a UTC receipt time.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations (readOnlyHint=false, idempotentHint=false, openWorldHint=true, destructiveHint=false) only sketch a profile; the description adds the genuinely non-obvious behavior that the upstream inventory list may auto-claim pending gifts and that claimed gifts come back with a UTC receipt time. That is exactly the side-effect detail an agent cannot get from annotations, and it stays consistent with destructiveHint=false.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences, front-loaded with the primary action and followed by the caveat/side effect. Every clause carries information: scope, exclusions, side effect, and return detail. No filler.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With no output schema present, the description partially compensates by noting that claimed gifts are returned with a UTC receipt time, and "rank" implies an ordered result. It stops short of describing the validation outcome shape (e.g., what a failed validation looks like), leaving a small gap for a tool whose whole purpose is producing a ranked, validated list.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The tool takes zero parameters, which is the baseline-4 case: there is nothing for the description to explain about arguments. No meaning is lost because the input schema is an empty object.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description names a specific verb pair ("Validate and rank") and resource ("cookable ledger recipes") with the data sources it draws on (live inventory and quota). It also explicitly rules out cooking/selling, which sharpens what the tool is. It does not, however, contrast itself with the sibling tools daily_routine or farm_brief, so the differentiation is implicit rather than stated.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Usage is implied: an agent can infer this is a pre-cooking validation step, and the "never cooks or sells" clause clarifies scope. But there is no explicit when-to-use or when-not-to-use guidance, no mention of alternatives, and no prerequisites or ordering relative to siblings.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.1.0daily_routinefarm_briefrecipe_checkScored across 3 tools
daily_routine is clearly the action-oriented tool, farm_brief is the read-only status tool, and recipe_check is the recipe-analysis tool. The main ambiguity is that daily_routine also optionally fetches recipe top five and updates the ledger, which slightly overlaps with recipe_check's recipe/ledger focus.
All names use lowercase snake_case and are readable, but they do not follow a consistent verb_noun or noun_verb pattern. daily_routine and farm_brief are noun phrases while recipe_check reads as an object-verb, so the convention is mixed though still understandable.
Three tools is at the low end of the ideal range but each tool covers a distinct area: daily execution, status reporting, and recipe validation. The count is slightly small because daily_routine bundles several unrelated actions, but it is still reasonable for a 'handy tools' server.
The set covers the main daily-routine, status, and recipe-checking workflows, but it lacks targeted action tools for many states that farm_brief reports (watering, chickens, bees) and explicitly cannot cook or sell recipes. This leaves notable dead ends, though the core workflow is usable.
Scraps Kitchen gives any AI agent a persistent, household-aware kitchen memory. Unlike generic chatbot recall, Scraps maintains structured cooking data: what's in your fridge (with freshness tracking), who you cook for (with allergens, dietary restrictions, and preferences), your recipe collection (with cook notes and per-diner ratings), your shopping list, and your kitchen equipment. 27 tools across 6 domains let agents read kitchen context, suggest meals that respect dietary safety, update the pantry after cooking, and build a history of what works for your household. Every interaction makes the data richer. Cooking history, preference signals, kitchen awareness = better suggestions next time. All tools work via oAuth and a free scraps.kitchen account.
Read-only Increase banking observability plus one safe non-money-moving write, for AI agents.
Read-only tools for agents: everything a Solana or Robinhood Chain wallet can get back (rent, wrapped SOL, LP fees, dead positions, failed deploys), whether a token is safe, when it unlocks, which protocol upgrades are live, and what a token operation costs. Every answer names the page where the user acts with their own wallet. Nothing here moves funds or holds keys.
Household-aware cooking brain: pantry, meal suggestions, dietary safety, recipes, shopping lists.
