workday-mcp-reference
A vendor-neutral reference implementation of a Workday MCP server that enables integration with Workday for employee data, compensation, benefits, time off, org info, inbox tasks, admin operations, and headcount management through natural language.
README
workday-mcp-reference
A vendor-neutral, clean-room reference implementation of a Workday MCP server. It reproduces the architecture of a production Workday Model Context Protocol server — OAuth broker, opaque session tokens with silent refresh, a REST + SOAP Workday client, regex SOAP parsing, text/HTML formatters, and per-tool gating — without any company-specific data. Use it as a starting point for your own Workday MCP integration.
Stack
- Python 3.11+, modern typing, absolute imports.
- MCP: official
mcpSDKFastMCP, streamable-HTTP transport. - HTTP:
requests(sharedSession+ retry on idempotent verbs only). - Logging: stdlib
logging. Metrics: a no-opMetricsprotocol (inject your own). - Tracing: optional OpenTelemetry, gated on
OTEL_EXPORTER_OTLP_ENDPOINT. - Config:
os.environ+python-dotenv.
Quick start
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]" # add ,otel,redis as needed
cp .env.example .env # fill in your tenant + OAuth client
workday-mcp-reference # serves streamable-HTTP on :8000/mcp
How auth works
- An MCP client calls
/mcpwith no valid bearer → the server returns401+WWW-Authenticatepointing at/.well-known/oauth-protected-resource. - The client discovers the authorization server (RFC 8414), optionally registers
(RFC 7591), then opens
/authorize→ the server 302s the user to Workday. - Workday redirects to
/callback?code=…; the server exchanges the code, resolves the user email, stores the Workday tokens under an opaque UUID, and returns that UUID. The client uses the UUID as its bearer thereafter. - Each tool call resolves the live Workday token from the store, refreshing silently when it is within 5 minutes of expiry.
Layout
src/workday_mcp_reference/
config.py environment-driven configuration
metrics.py Metrics protocol + NoopMetrics
server.py FastMCP + Bearer gate + OAuth/health routes + main()
auth/
token_store.py opaque-id → Workday tokens, silent refresh
oauth.py authorize/callback broker + discovery metadata
workday/
client.py WorkdayClient (REST + SOAP), uniform envelope
parsers.py regex SOAP parsing (no lxml)
formatters.py text + optional HTML artifacts
tools/
__init__.py register_all + get_client/resolve_me_wid helpers
me.py exemplar tool module (workday_me)
pay.py compensation + one-time payment (REST + SOAP)
benefits.py enrollments, elections, total rewards
time.py time off: scheduled, balance, eligible types, request
org.py manager, reports, org chart, search, lookup
inbox.py inbox tasks: list, approve, deny
admin.py job profiles, pay groups, hire steps, BP status
headcount.py open positions, approvals, position detail (RaaS/WQL)
The eight category tools each take an action argument and dispatch to one
implementation — mirroring how a production server keeps the MCP tool surface
small. headcount.py depends on tenant-specific RaaS reports and WQL fields, so
its report names and field mappings are driven entirely by env config (see
.env.example); the other modules use standard Workday REST/SOAP APIs.
Adding a tool module
Copy tools/me.py. Expose def register(mcp: FastMCP) -> None, define your
@mcp.tool() functions inside it, and call get_client() / resolve_me_wid(client)
— never read tokens directly. Add your module name to _TOOL_MODULES in
tools/__init__.py.
推荐服务器
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。
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。
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。
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。
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。
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。
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。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
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。