Open Source · SAP Cloud ALM MCP Server · Version 1.0
Connect Claude, GitHub Copilot, Cursor or Microsoft Copilot Studio to SAP Cloud ALM. Ask about requirements, user stories, defects, tests and processes in plain language, and have documentation, user stories and links created right inside Cloud ALM.
Overview
calmcp is a CALM MCP server: an MCP server for SAP Cloud ALM, CALM for short. It turns your AI assistant's questions into calls to the official Cloud ALM APIs and hands the answers back in a structured form.
The Model Context Protocol (MCP) is the open standard AI assistants use to reach external systems. As a Cloud ALM MCP server, calmcp makes SAP Cloud ALM one of those systems: project leads, test managers and developers ask in chat instead of filtering lists, building exports and adding up numbers by hand.
calmcp runs locally on your machine or centrally in your own SAP BTP subaccount. There is no Consetto service in between: your data goes straight from Cloud ALM to your AI client.
How it feels
Ask the way you would ask a colleague. calmcp picks the right Cloud ALM API, counts on the server and gives the assistant exactly the data it needs for the answer.
count_only and group_by return real totals.complete: false instead of passing off a partial number as the total.Example conversation with made-up data.
Features
Instead of dozens of single-purpose tools, calmcp bundles the Cloud ALM APIs into six clear, intent-based tools. That keeps your assistant's context lean and its answers precise.
calm_list
Tasks, requirements, user stories, defects, projects, features, transports, documents, test cases, hierarchy nodes and library objects. With OData filters, field projection, sprint filter, uniform paging and live counts.
calm_get
Fetches a single record by ID, a feature also by its display ID such as 6-123. Ideal for follow-up questions in the conversation.
calm_analytics
Queries the Cloud ALM analytics providers (Defects, Tasks, Tests …) and aggregates across projects. It counts records, not analytics rows.
calm_resources
A catalogue of every resource, the status, type and priority codes, and ready-made recipes. The assistant works out on its own how to answer a question.
calm_create
Optional: creates documents, features, tasks (user stories, defects, requirements), comments and library entries, and links existing objects: tasks to features, documents, processes and other tasks, test cases to user stories. Name existing objects by display id. Never deletes.
calm_update
Optional: changes fields of existing features such as title, description, status and priority. If the feature changed in the meantime, calmcp refuses, and images in the description are never lost unnoticed.
stdio · HTTP · BTP
As a one-click extension in Claude Desktop, locally for Claude Code, Cursor and VS Code, or centrally on SAP BTP Cloud Foundry with single sign-on for the whole team.
Use cases
Copy a question, paste it into your assistant and see for yourself.
Status reports, stand-up digests and steering committee decks in minutes rather than hours.
Defect status, test coverage and test plans at a glance, across projects.
From the business process to the open work on it, in both directions.
Understand the requirement before writing the first line of code, and keep your developments properly inventoried.
With write access: create specifications as Cloud ALM documents and link them to tasks and features.
Who sits where, in which role, across several projects.
Click a question to copy it to your clipboard.
Installation
Pick the path that suits you. Every option runs the same server with the same tools.
A service key for the SAP Cloud ALM API from your BTP subaccount (OAuth2 client credentials: tenant, region, client ID, client secret). The region is for example eu10, eu20, us10 or ap10.
Use the sandbox of the SAP Business Accelerator Hub with your personal API key. No tenant of your own required.
The simplest path for individuals. calmcp ships as a Claude Desktop extension (.mcpb) and runs on macOS, Windows and Linux.
calmcp-<version>.mcpb from the GitHub releases.Write and update access stay off until you switch on Write Access or Update Access in the extension settings.
Requires Node.js 22 or later. Build once, then register it in any MCP client.
git clone https://github.com/consetto/calmcp.git
cd calmcp && npm install && npm run build
claude mcp add calmcp \
-e CALM_TENANT=<tenant> -e CALM_REGION=eu10 \
-e CALM_CLIENT_ID=<client-id> -e CALM_CLIENT_SECRET=<secret> \
-- node "$PWD/dist/index.js"Cursor, Claude Desktop and other clients use the same configuration as JSON (in VS Code the key is servers instead of mcpServers):
{
"mcpServers": {
"calmcp": {
"command": "node",
"args": ["/absolute/path/to/calmcp/dist/index.js"],
"env": {
"CALM_TENANT": "<tenant>",
"CALM_REGION": "eu10",
"CALM_CLIENT_ID": "<client-id>",
"CALM_CLIENT_SECRET": "<secret>"
}
}
}
}Sandbox instead of a tenant: set "CALM_SANDBOX": "true" and "CALM_API_KEY": "<key>".
For teams: deploy once, and everyone signs in with their BTP user. The MTA creates XSUAA, the destination service and application logging for you.
git clone https://github.com/consetto/calmcp.git
cd calmcp && npm install
mbt build
cf deploy mta_archives/calmcp_1.0.0.mtarSAP_CALM, type HTTP, URL https://<tenant>.<region>.alm.cloud.sap/api, authentication OAuth2ClientCredentials with the values from your Cloud ALM service key.CALMCP_Viewer role collection to your users, and CALMCP_Editor to those who may create and update.https://<route>/mcp in your AI client. Sign-in runs interactively through XSUAA; calmcp never sees a password. On first sign-in you approve the client once.claude mcp add --transport http calmcp https://<route>/mcpIn Claude Desktop: Settings → Connectors → custom connector with the /mcp URL. The full guide is in the README. Upgrading an existing deployment: run a full cf deploy, see the OAuth upgrade guide. A few parameters were renamed from 0.9 to 1.0; the list is in the changelog.
Copilot Studio calls server-to-server, so it uses an API key instead of a user sign-in. It builds on the BTP deployment from the previous tab.
cf set-env calmcp-srv CALM_HTTP_API_KEY "$(openssl rand -base64 48)", then cf restart calmcp-srv.copilot-studio/calmcp-connector.json from the repository and set its host to your route.Bearer <your-key> as the key.Step by step: Copilot Studio guide on GitHub.
Security & trust
An AI assistant that reads your project memory has to follow clear rules. calmcp makes those rules the default.
Without explicit opt-in calmcp only reads. Nothing is ever deleted.
Create and update are separate switches, available only to users with CALMCP_Editor, strictly validated against the OpenAPI specification. Every change is written to an audit log.
XSUAA with OAuth and dynamic client registration, client approval at sign-in, redirects only to known MCP clients, the CALMCP_Viewer and CALMCP_Editor role collections, or an API key for server clients.
If Cloud ALM leaves a write unanswered, nothing is retried and calmcp says the outcome is unknown. Rich text is checked against an allowlist.
Semantic versioning: tools, parameters, error codes, scopes and roles only change with a new major version. Every change is in the changelog.
Runs on your machine or in your BTP subaccount. No Consetto service in between, source code open under MIT.
FAQ
calmcp is an open-source CALM MCP server, i.e. an MCP server for SAP Cloud ALM (CALM for short). It exposes the Cloud ALM APIs through the Model Context Protocol so that AI assistants such as Claude, GitHub Copilot, Cursor or Microsoft Copilot Studio can query requirements, user stories, defects, tests, processes and documents in plain language.
Nothing. calmcp is released under the MIT license and freely available on GitHub. All you need is access to SAP Cloud ALM and an AI client that supports MCP.
Delete, never. calmcp is read-only by default. With CALM_WRITE_ENABLED=true an administrator can additionally allow creating new documents, features, tasks, comments and library entries and linking existing objects, and with CALM_UPDATE_ENABLED=true changing fields of existing features, guarded against overwriting concurrent edits.
Yes. From version 1.0 on, calmcp follows semantic versioning: tool and parameter names, error codes, the shape of counts and write answers, environment variables, scopes and role collections only change with a new major version. Every change is listed in the changelog.
Any MCP-capable client, for example Claude Desktop, Claude Code, Cursor, VS Code with GitHub Copilot and Microsoft Copilot Studio. Locally calmcp runs over stdio; for teams it runs over Streamable HTTP on SAP BTP.
An SAP Cloud ALM tenant with a service key for the Cloud ALM API (OAuth2 client credentials), or, to try it out, an API key for the SAP Business Accelerator Hub sandbox. Installing from source requires Node.js 22 or later.
Tasks (including requirements, user stories, defects, risks and quality gates), projects and programs, features, documents, process hierarchy, process scopes, custom processes, test management, test plans, analytics, status events, landscape and the cross-library with applications, configurations, developments and interfaces.
calmcp runs on your machine or in your own SAP BTP subaccount and calls the Cloud ALM APIs directly. Consetto does not operate any intermediate service. Which data is sent to a language model is decided by the AI client you choose.
Install calmcp, give the project a star on GitHub and tell us what you build with it. For team rollouts, BTP deployment or a workshop, we are happy to help.