YAZIO MCP

YAZIO MCP

Enables AI agents to interact with a personal YAZIO nutrition diary, including searching products, reading and writing diary entries, and analyzing nutrition trends.

Category
访问服务器

README

YAZIO MCP

English · Русский

Node.js 20+ TypeScript MCP 2026-07-28 License: MIT

A local Model Context Protocol server for working with your personal YAZIO nutrition diary from AI agents such as Claude Code, Claude Desktop, Codex, and Hermes Agent.

yazio-mcp speaks MCP over stdio and talks directly to YAZIO through a small, typed adapter maintained in this repository. It does not depend on the legacy yazio npm package or on a third-party proxy service.

[!WARNING] YAZIO does not publish or support a public developer API. This project uses a reverse-engineered, unofficial API that may change or be blocked without notice. Use it only with your own account and at your own risk. This project is not affiliated with, endorsed by, or sponsored by YAZIO GmbH.

What you can do

  • Search the YAZIO product database using international English defaults (US, en_US) or configure any supported country and locale.
  • Read breakfast, lunch, dinner, and snack entries for one day or a date range.
  • Inspect daily calories, macros, water, weight, goals, and exercises.
  • Add individual products or a compensated batch of meal items.
  • Change the amount of an existing product entry, for example 200 g → 210 g.
  • Remove an exact diary entry after explicit confirmation.
  • Add water using a guarded read-modify-write flow.
  • Compare periods and analyze nutrition trends by day, ISO week, or month.
  • Flag days or meal slots that may be missing entries without inventing food.

Every tool returns:

  • an outputSchema;
  • machine-readable structuredContent;
  • a short text fallback for older clients;
  • normalized calories, protein, fat, carbohydrates, fiber, and sugar;
  • a consistent error envelope with stable error codes.

Safety model

Nutrition writes are deliberately conservative:

  • dry_run defaults to true;
  • every write requires a client_request_id;
  • duplicate detection runs before product writes;
  • results are verified by reading the diary again;
  • idempotency state survives process restarts;
  • destructive actions use MCP multi-round-trip confirmation when supported;
  • legacy clients receive a short-lived, one-time confirmation token;
  • ambiguous writes are never retried blindly.

YAZIO does not expose a confirmed PATCH endpoint for consumed items. update_consumed_item therefore performs a guarded replacement:

  1. Read and identify the exact original entry.
  2. Add a replacement with a new UUID.
  3. Verify that exact replacement UUID.
  4. Delete the original UUID.
  5. Verify the final state.
  6. Remove the replacement as compensation if the original could not be deleted safely.

Updates of the same diary item are serialized across processes that share the same state file.

Available tools

Read tools

Tool Description
get_user_profile Privacy-filtered profile without email, birth date, tokens, or payment identifiers
get_goals Calorie, macro, water, step, and weight goals
search_products Text search with configurable country and locale
get_product Product details and normalized nutrients per 100 g/ml
get_diary Diary for one date or a range of up to 31 days
get_daily_summary Daily nutrition, goals, water, steps, and four meal buckets
get_water_intake Cumulative water intake for a date
get_weight Latest known weight on or before a date
get_exercises Training and custom training entries

Write tools

Tool Description
add_consumed_item Add one product with preview, duplicate detection, idempotency, and verification
add_meal_batch Add up to 25 products and compensate earlier writes after a partial failure
update_consumed_item Safely replace the amount of an existing product entry
remove_consumed_item Remove an exact diary item after MRTR or one-time-token confirmation
add_water_intake Add water by updating the cumulative daily total

Analytics tools

Tool Description
get_nutrition_trends Day/week/month averages, goals, variance, and stability
compare_periods Compare calorie and macro averages for two periods
detect_probable_missing_meals Report suspiciously empty meal slots; never writes food automatically

Requirements

  • Node.js 20 or newer
  • npm
  • A personal YAZIO account that can sign in with email and password

Node.js 22 or 24 LTS is recommended for regular use.

Quick start

git clone https://github.com/tomastaker/yazio-mcp.git
cd yazio-mcp
npm ci
npm run build

The server entry point is:

node /absolute/path/to/yazio-mcp/dist/index.js

The process uses stdout exclusively for MCP JSON-RPC. Transport diagnostics are written to stderr.

Authentication and configuration

The server does not automatically load .env files. Pass credentials through your MCP client's environment configuration or a local secret manager.

Required for the first login:

Variable Description
YAZIO_USERNAME Email address of your personal YAZIO account
YAZIO_PASSWORD Password of your personal YAZIO account

After login, the access and refresh tokens are cached locally. The password is never written to the token cache. A cached token is bound to a SHA-256 fingerprint of the normalized username, API base URL, and OAuth client ID, which prevents accidentally reusing a token from another account.

YAZIO does not offer public OAuth application registration. The adapter includes community-observed application-level OAuth credentials published by the MIT-licensed saganos/yazio_public_api project. They are shared application identifiers, not a user's personal secret. Both can be overridden together if YAZIO rotates them.

Optional variables

Variable Default Purpose
YAZIO_BASE_URL https://yzapi.yazio.com/v19 Trusted local API configuration
YAZIO_TOKEN_CACHE_PATH ~/.yazio-mcp/token.json Access/refresh token cache
YAZIO_MCP_STATE_PATH ~/.yazio-mcp/state.json Idempotency and confirmation state
YAZIO_MCP_TIMEZONE UTC IANA timezone used for dates
YAZIO_PRODUCT_COUNTRIES US Comma-separated product countries
YAZIO_PRODUCT_LOCALES en_US Comma-separated product locales
YAZIO_DEFAULT_SEX male Search parameter required by the upstream API
YAZIO_HTTP_TIMEOUT_MS 15000 HTTP timeout, from 1,000 to 120,000 ms
YAZIO_CLIENT_ID built in Advanced paired OAuth override
YAZIO_CLIENT_SECRET built in Advanced paired OAuth override

Country, locale, and timezone are independent. For example, a German setup can use YAZIO_PRODUCT_COUNTRIES=DE, YAZIO_PRODUCT_LOCALES=de_DE, and YAZIO_MCP_TIMEZONE=Europe/Berlin. Multiple search values can be supplied as comma-separated lists; explicit countries and locales tool arguments override these defaults for an individual search.

NUTRITION_MCP_STATE_PATH and NUTRITION_MCP_TIMEZONE are accepted as deprecated migration aliases. Conflicting old and new values are rejected.

Client configuration

Replace /absolute/path/to/yazio-mcp in the examples below with the real checkout path. Prefer environment forwarding or an OS secret manager over storing credentials directly in a client configuration file.

Claude Code

Create .mcp.json in your project:

{
  "mcpServers": {
    "yazio": {
      "type": "stdio",
      "command": "node",
      "args": ["/absolute/path/to/yazio-mcp/dist/index.js"],
      "env": {
        "YAZIO_USERNAME": "${YAZIO_USERNAME}",
        "YAZIO_PASSWORD": "${YAZIO_PASSWORD}"
      }
    }
  }
}

Verify the connection:

claude mcp list

Claude Desktop

Add the server to:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "yazio": {
      "command": "node",
      "args": ["/absolute/path/to/yazio-mcp/dist/index.js"],
      "env": {
        "YAZIO_USERNAME": "set-locally",
        "YAZIO_PASSWORD": "set-locally"
      }
    }
  }
}

Claude Desktop stores literal environment values in its configuration. Protect that file with appropriate OS permissions.

Codex

Forward environment variables from ~/.codex/config.toml:

[mcp_servers.yazio]
command = "node"
args = ["/absolute/path/to/yazio-mcp/dist/index.js"]
env_vars = [
  "YAZIO_USERNAME",
  "YAZIO_PASSWORD"
]
default_tools_approval_mode = "writes"
startup_timeout_sec = 15
tool_timeout_sec = 90
enabled = true

Or register it with the CLI:

codex mcp add yazio \
  --env YAZIO_USERNAME="$YAZIO_USERNAME" \
  --env YAZIO_PASSWORD="$YAZIO_PASSWORD" \
  -- node /absolute/path/to/yazio-mcp/dist/index.js

Environment forwarding is safer than the CLI variant because literal --env values may be saved in local client configuration.

Hermes Agent

Add the server to ~/.hermes/config.yaml:

mcp_servers:
  yazio:
    command: "node"
    args: ["/absolute/path/to/yazio-mcp/dist/index.js"]
    env:
      YAZIO_USERNAME: "set-locally"
      YAZIO_PASSWORD: "set-locally"
      YAZIO_MCP_TIMEZONE: "UTC"
    timeout: 90
    connect_timeout: 15
    supports_parallel_tool_calls: false
    enabled: true

Parallel tool calls are disabled here as a conservative default for an unofficial API.

Example prompts

Search the YAZIO product database for Snickers and show me the best matches.

Show everything I logged for breakfast, lunch, dinner, and snacks today.

First show a dry-run, then add 80 g of product <product-uuid> to breakfast.

Change diary item <item-uuid> from 200 g to 210 g, but show me the preview first.

Compare my average calories and macros for the last two complete weeks.

Find days with probably missing breakfasts, but do not add anything.

An agent normally searches first, selects a product UUID, previews the write, and only then submits the mutation.

Responses and errors

Successful tools return both text and structured data:

{
  "ok": true,
  "data": {},
  "meta": {
    "timezone": "UTC",
    "backend": "yazio-unofficial",
    "generated_at": "2026-07-30T00:00:00.000Z"
  }
}

Failures set isError: true and use a machine-readable code:

{
  "ok": false,
  "error": {
    "code": "DUPLICATE_DETECTED",
    "message": "A probable duplicate entry was found; nothing was written.",
    "retryable": false
  }
}

For diary rows whose individual nutrition cannot be determined safely, get_diary exposes:

  • totals_complete;
  • totals_source (item_sum or daily_summary);
  • missing_nutrition_item_ids.

This prevents a partial item sum from being presented as a complete total.

Recipes and unsupported operations

The adapter can read user recipe IDs and recipe details using community-observed endpoints. Recipe rows in the diary preserve recipe_id, and their names are enriched on a best-effort basis.

The following operations are intentionally not exposed as successful MCP tools because their upstream contracts are not sufficiently confirmed:

  • adding a saved recipe portion to the diary;
  • creating, updating, or deleting recipes;
  • changing a recipe portion through update_consumed_item;
  • native PATCH/PUT of consumed items;
  • native date-range endpoints;
  • full weight history;
  • native barcode lookup;
  • guaranteed transactional meal batches.

Recipe-row nutrients remain null when the upstream basis cannot be determined safely. Daily totals use the daily-summary endpoint instead.

Development

npm run lint
npm run typecheck
npm test
npm run build
npm run smoke

Run all default checks:

npm run check

The protocol smoke test starts a local mock YAZIO server, spawns the built MCP server over stdio, and validates:

  • modern MCP 2026-07-28 negotiation;
  • a real MRTR update/remove flow;
  • legacy one-time-token confirmation;
  • tool input/output schemas;
  • structuredContent and privacy filtering.

Live tests

Live tests are opt-in:

export YAZIO_LIVE_TEST=1
export YAZIO_USERNAME='...'
export YAZIO_PASSWORD='...'
npm run test:live

The live write test creates a minimal product entry and removes that exact generated UUID in finally. If cleanup cannot be confirmed, the test fails. Avoid running live writes on an important diary date without checking the result in the YAZIO app.

Privacy and security

  • The server is designed for one local user, not as a hosted multi-user API.
  • User credentials are accepted only from environment variables.
  • Passwords, authorization headers, tokens, and full profile responses are not written to logs or surfaced in tool errors.
  • Token and state files use atomic writes and POSIX mode 0600; the default directory is created with 0700.
  • Windows users must protect the cache directory with Windows ACLs because POSIX mode bits are not an ACL substitute.
  • The profile tool uses an explicit allowlist and excludes email, birth date, payment identifiers, and upstream tokens.
  • One-time confirmation tokens are stored only as SHA-256 digests.
  • Write requests are not retried blindly after uncertain network failures.

See SECURITY_REVIEW.md for the current threat review.

API and MCP compatibility

The default /v19 API base was selected from observed behavior. Newer-looking URL versions are not necessarily released API versions; YAZIO can block or change any of them without warning.

The project uses the official split TypeScript SDK packages:

  • @modelcontextprotocol/server@2;
  • @modelcontextprotocol/client@2 for protocol tests.

The server opts into MCP 2026-07-28 through serveStdio and keeps the legacy stdio compatibility shim enabled for existing clients.

References:

Implementation provenance and license notices are documented in THIRD_PARTY_NOTICES.md.

License

MIT

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选