Cloud ALM MCP
Enables local or HTTP-based MCP interactions with a mock SAP Cloud ALM environment, enforcing read/write capability separation and fail-closed security policies for future SAP Cloud ALM connectivity.
README
Cloud ALM MCP
Node.js and TypeScript skeleton for a SAP Cloud ALM Model Context Protocol server.
This repository is currently mock-first. It can install, build, test, and start without SAP Cloud ALM credentials, BTP Destination Service configuration, or OAuth setup.
Target Architecture
- Runtime: Node.js on SAP BTP Cloud Foundry.
- Production MCP transport: MCP Streamable HTTP via the official MCP TypeScript SDK.
- STR-158 uses stateless Streamable HTTP for the POC by setting no MCP session ID generator.
- Local fallback transport: stdio, intended only for local development.
- Future Cloud ALM access path: SAP BTP Destination Service.
- Future authentication model: OAuth2 Client Credentials through a configured destination.
Real SAP Cloud ALM connectivity is intentionally not included in STR-158.
Session and state requirements must be revisited before production agent usage if later tool flows require persistent MCP session state. STR-158 intentionally does not introduce a session store.
Local Development
npm install
npm run build
npm test
npm start
By default, npm start runs mock mode over HTTP and exposes:
- MCP endpoint:
POST /mcp - Health endpoint:
GET /health
The remote MCP endpoint is stateless and POST-only. GET, DELETE, and other unsupported methods on /mcp return 405 Method Not Allowed with Allow: POST; GET SSE and MCP session termination are intentionally not implemented. JSON request bodies are limited to 64kb.
Example:
curl http://localhost:3000/health
Use .env.example as the list of supported placeholders. Do not add real Cloud ALM secrets to local files committed to git.
SAP BTP Cloud Foundry POC Deployment
manifest.yml defines a single cloud-alm-mcp Cloud Foundry application using nodejs_buildpack, command: npm start, and an HTTP health check on /health. The first deployment is intentionally locked to safe mock mode:
RUNTIME_MODE=mock
MCP_TRANSPORT=http
EXTERNAL_CALLS_ENABLED=false
READ_CAPABILITY_ENABLED=true
WRITE_CAPABILITY_ENABLED=false
No Cloud ALM credentials, OAuth settings, XSUAA/IAS binding, Destination Service binding, or real destination values are required for STR-162.
Cloud Foundry staging runs npm install; the package postinstall lifecycle runs npm run build so dist/src/index.js exists before npm start. TypeScript and the type packages needed for compilation are regular dependencies so the standard Node.js buildpack production install can build the app without carrying local-only tools such as vitest and tsx.
The app keeps using the platform-provided PORT; do not hardcode a production port. package.json requests Node.js 22.x for deterministic staging/runtime behavior on the first deployment. SAP BTP Cloud Foundry currently supports Node.js 22 through nodejs_buildpack; operators should still verify the exact target foundation with cf buildpacks before deployment.
package-lock.json remains the authoritative npm dependency lock. No packageManager field is set because Cloud Foundry uses npm for a root package.json/package-lock.json, and forcing an npm version is only needed if the target foundation's default npm proves incompatible.
Local validation:
npm install
npm run build
npm test
npm start
curl http://localhost:3000/health
BTP values needed from the operator:
CF API endpoint
BTP org
BTP space
route/domain decision, if not using the default route
Manual BTP validation flow:
cf login -a <api-endpoint>
cf target -o <org> -s <space>
cf buildpacks
cf push
cf app cloud-alm-mcp
cf logs cloud-alm-mcp --recent
curl https://<route>/health
After deployment, validate that /health returns status: ok, POST /mcp works in mock mode, GET /mcp returns 405 Method Not Allowed with Allow: POST, and logs contain request IDs, method, path, status, and duration without request bodies, authorization headers, tokens, or secrets.
Configuration
Core environment variables:
RUNTIME_MODE=mock|destinationMCP_TRANSPORT=http|stdioPORT=3000READ_CAPABILITY_ENABLED=true|falseWRITE_CAPABILITY_ENABLED=true|falseALLOWED_DESTINATIONS=cloud-alm-devEXTERNAL_CALLS_ENABLED=false|true
Local defaults are intentionally safe: mock runtime, HTTP transport, read enabled, write disabled, and external calls disabled.
Destination mode exists only as a placeholder in this skeleton. It fails closed until the BTP Destination Service lookup and OAuth token flow are implemented in later tasks.
Mock Task Client
Mock mode uses deterministic in-memory task and comment data for local development and tests. It supports only the current mock contract fields: task id, title, status, priority, and comment id, taskId, author, text, createdAt.
The mock client returns known tasks, rejects unknown task IDs, appends one comment at a time with deterministic IDs, and updates only explicitly allowlisted mock task fields. These fields are not claimed to be official SAP CALM_TKM payload fields. Real Cloud ALM endpoints, payload schemas, scopes, pagination names, and update semantics remain unverified and deferred to the real integration work.
Security Boundary
The Policy Guard is code-enforced before Cloud ALM client calls. It validates:
- allowed operation names only,
- read/write capability separation,
- delete operation rejection,
- bulk operation rejection,
- unknown operation rejection,
- no agent-supplied destination or customer selection,
- fail-closed behavior for ambiguous configuration.
This is deliberately not prompt-only enforcement. Customer isolation and durable audit logging are left as future architecture work, with module boundaries already present.
Audit events already reserve optional fields for future traceability: actor, customer context, resource type/id, and correlation ID. The skeleton does not invent real actor or customer values and does not log request payloads, tokens, authorization headers, client IDs, client secrets, or sensitive response bodies.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.