@mcp-abap-adt/calm-server
MCP server for SAP Cloud ALM, providing 54 tools across 9 services to manage features, tasks, test cases, documents, projects, and more via natural language.
README
@mcp-abap-adt/calm-server
MCP server for SAP Cloud ALM, built on
@mcp-abap-adt/calm-client.
Ships 54 MCP tools covering all 9 Cloud ALM services — full CRUD
where the service supports it, plus a wide read surface — with rich
JSON Schema descriptions that let an LLM plan multi-step workflows.
This package is dual-purpose:
- Runnable stdio server —
npx @mcp-abap-adt/calm-server. Plug directly into Claude Desktop, Claude Code, or any MCP-compatible host. - Reusable library of tool primitives — import individual tools
(
FEATURES_GROUP,TASKS_GROUP, …) and embed them in a larger composed MCP server without reimplementing anything.
Status
0.2.0 — 54 tools, 232 tests (224 unit + integration, 7 env-gated
skips, 1 todo), full build green. Integration suite runs live against
the SAP sandbox (api.sap.com) or any OAuth2 Cloud ALM tenant with a
single .env switch; gates skip cleanly when no backend is wired.
Installation
As a standalone MCP server
npm install -g @mcp-abap-adt/calm-server
# or per-project:
npm install @mcp-abap-adt/calm-server
As a library (compose into your own MCP server)
npm install @mcp-abap-adt/calm-server
# peers:
npm install @mcp-abap-adt/calm-client @mcp-abap-adt/interfaces @modelcontextprotocol/sdk
Standalone: running the server
1. Configure credentials
Copy the template and fill in:
cp .env.example .env
OAuth2 mode (real tenant) — paste values from an XSUAA service key:
CALM_MODE=oauth2
CALM_BASE_URL=https://<tenant>.<region>.alm.cloud.sap
CALM_UAA_URL=https://<tenant>.authentication.<region>.hana.ondemand.com
CALM_UAA_CLIENT_ID=sb-…!b…|calm!b…
CALM_UAA_CLIENT_SECRET=…
Sandbox mode (SAP API Business Hub):
CALM_MODE=sandbox
CALM_API_KEY=<your-key-from-api.sap.com>
# CALM_BASE_URL defaults to https://sandbox.api.sap.com/SAPCALM
2. Launch
# Global install:
calm-mcp
# Or via npx without install:
npx @mcp-abap-adt/calm-server
# Or from a clone:
npm run build && node dist/bin/stdio.js
The server speaks MCP over stdio. Misconfiguration is reported to
stderr with a non-zero exit code.
Authentication setup
calm-mcp supports two OAuth2 flows for live tenants, selectable via
CALM_AUTH_FLOW:
| Flow | Use case | Browser? | Refresh token? |
|---|---|---|---|
client_credentials (default) |
technical service-binding (sb-* client) |
no | no |
authorization_code |
end-user dev workflow, full user scope | once | yes |
Option A — quick CC setup (technical user)
Plain .env with inline CALM_UAA_URL / CALM_UAA_CLIENT_ID /
CALM_UAA_CLIENT_SECRET works as before — no extra steps. The broker uses
an in-memory session shim for these inline creds.
Option B — broker-backed setup (CC or AC)
Use the bundled mcp-auth
CLI to convert a BTP service key into a token-bearing .env:
# CC (no browser)
npx mcp-auth --service-key ./sk.json --output ./DEFAULT.env \
--type xsuaa --credential
# AC (browser pops once; refresh_token persists)
npx mcp-auth --service-key ./sk.json --output ./DEFAULT.env \
--type xsuaa --browser auto
Then in your .env:
CALM_MODE=oauth2
CALM_BASE_URL=https://<tenant>.<region>.alm.cloud.sap
CALM_AUTH_FLOW=authorization_code # or client_credentials
CALM_DESTINATION=DEFAULT
The server's runtime auth pipeline is @mcp-abap-adt/auth-broker.
Note:
buildCalmClientis async since v0.4.0 (was sync in v0.3.x). Library consumers mustawaitit.
3. Wire into Claude Desktop
Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json
on macOS, %APPDATA%\Claude\claude_desktop_config.json on Windows):
{
"mcpServers": {
"calm": {
"command": "npx",
"args": ["-y", "@mcp-abap-adt/calm-server"],
"env": {
"CALM_MODE": "sandbox",
"CALM_API_KEY": "<your-sandbox-key>"
}
}
}
}
Restart Claude Desktop; the 54 calm_* tools become available to the
model.
Library: composing into another MCP server
Useful when you want to expose Cloud ALM tools alongside ADT tools, Reports tools, or your own domain tools in a single MCP process.
import { CalmClient, CalmConnection } from '@mcp-abap-adt/calm-client';
import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
import {
ALL_GROUPS,
BaseCalmMcpServer,
CalmToolRegistry,
FEATURES_GROUP,
TASKS_GROUP,
} from '@mcp-abap-adt/calm-server';
// Option A — BaseCalmMcpServer with a curated subset
const calm = new CalmClient(
new CalmConnection({ baseUrl, apiKey }),
);
const server = new BaseCalmMcpServer({
name: 'my-mcp',
version: '1.0.0',
calm,
groups: [FEATURES_GROUP, TASKS_GROUP], // only these land as tools
});
// Option B — embed into your existing McpServer
const existing = new McpServer({ name: 'combined', version: '1.0.0' });
const registry = new CalmToolRegistry([...ALL_GROUPS]);
registry.registerAll(existing, () => ({ calm }));
// Now `existing` serves Cloud ALM tools + whatever else you registered.
Subpath exports:
import { ALL_GROUPS } from '@mcp-abap-adt/calm-server/tools';
import { CalmToolRegistry } from '@mcp-abap-adt/calm-server/registry';
Tool surface (54 tools across 9 services)
| Service | Tools |
|---|---|
| Features (11) | list, get, get_by_display_id, create, update, delete, create_external_reference, delete_external_reference, list_external_references, list_statuses, list_priorities |
| Tasks (10) | list, get, create, update, delete, list_comments, create_comment, list_references, list_deliverables, list_workstreams |
| TestCases (9) | list, get, create, update, delete, create_activity, create_action, list_activities, list_actions |
| Documents (7) | list, get, create, update, delete, list_statuses, list_types |
| Projects (7) | list, get, create, list_programs, get_program, list_team_members, list_timeboxes |
| Hierarchy (5) | list, get_with_children, create_node, update_node, delete_node |
| Analytics (2, read-only) | query (17 endpoints), list_providers (static catalog) |
| Logs (2, domain-specific REST) | get (provider + serviceId + time window), post (inbound log records) |
| ProcessMonitoring (1, read-only) | list_processes |
Every MCP tool:
- Has a full JSON Schema with descriptions — the LLM reads these to plan.
- Wraps arguments into OData
$filter/$select/$top/$skipinternally — the LLM never sees raw OData syntax. - Returns compact records by default (
limit=20,fieldsdefault ≈ 6 columns per entity); callers opt into more viafields,limit,withCount,offset. - Maps
CalmApiError→ a typed MCP error the LLM can branch on (NOT_FOUND,NETWORK,ODATA_ERRORwithserviceCode, …).
See src/tools/<service>/*.ts for per-tool argument schemas.
Destructive tools (write operations)
Every Cloud ALM service that supports writes is now exposed:
- Features:
create,update,delete, plus external-referencecreate/delete - Tasks:
create,update,delete,create_comment - TestCases:
create,update,delete, plus nestedcreate_activity/create_action - Documents:
create,update,delete - Hierarchy:
create_node,update_node,delete_node - Projects:
create - Logs:
post(inbound OpenTelemetry-style record ingestion)
The shared SAP sandbox at api.sap.com is read-friendly only — mutation
integration tests are opt-in (see Live-tenant integration below).
Debug logging
CALM_LOG_LEVEL=debug # error | warn | info | debug
DEBUG_CALM_CONNECTORS=true # CalmConnection retries, 401 refresh, URLs
DEBUG_CALM_LIBS=true # resource-client internals
DEBUG_CALM_TESTS=true # test execution progress
Logging goes to stderr (stdout is reserved for the MCP protocol stream).
Development
git clone git@github.com:fr0ster/mcp-calm-server.git
cd mcp-calm-server
npm install
npm run test # 224 unit + integration tests
npm run build # emits dist/, includes executable bin
npm run lint:check # biome
Live-tenant integration
src/__tests__/integration/ runs against a real backend when env is
present, and skips cleanly when it isn't (so npm test without secrets
is always green). Five gates, drop them into .env:
| Gate | Env trigger | What it unlocks |
|---|---|---|
describeSandbox |
CALM_MODE=sandbox + CALM_API_KEY |
The api.sap.com sandbox |
describeOAuth2 |
CALM_MODE=oauth2 + CALM_BASE_URL + 3× UAA env |
Live tenant (incl. endpoints absent from the sandbox catalog, e.g. Business Processes) |
describeWhenLive |
either of the above | Read-side tests that work in either mode |
describeWithProject |
live backend + CALM_PROJECT_ID |
Project-scoped chains (features list→get, tasks list→get→comments, …) |
describeMutating |
live + CALM_PROJECT_ID + CALM_ALLOW_MUTATIONS=1 |
Write tests (every mutation finalises via try/finally { delete }) |
A quick smoke script lives at scripts/smoke-sandbox.mjs — spawns the
stdio bin, lists tools, calls a handful of read endpoints, and exits
1 on any non-skip failure.
License
MIT — see LICENSE.
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。