cap-add-remote-service

What I do

Guide the developer step-by-step through adding a remote service integration to an existing CAP Node.js project, following the CAP-Level Service Integration (Calesi) pattern. This covers importing APIs, defining consumption views, integrating the model, and wiring up the required Node.js handlers.

Context I need first

Before starting, inspect the project so I understand its structure:

1. Read package.json to learn the project name and existing dependencies.
2. Use cds-mcp_search_model to get an overview of the services and entities already defined in the project.
3. Ask the user:
   • Which remote service they want to integrate (another CAP app, S/4HANA Business Partner, etc.)
   • Whether they have the API already as an npm package, an .edmx file, or just a URL

Step 1 — Import the API

Option A: CAP package from npm

If the remote service is published as a CAP package (e.g. @capire/xflights-data):

npm add <package-name>

Check the added package's index.cds and package.json to understand what service name and entities it exposes.

Important: Also check the package's mock data CSV files (usually in srv/external/data/) to see which columns actually have data. The consumption view should only project columns that exist in the mock data, otherwise the UI will show empty values during local development. (Check back with the user if unsure)

Option B: OData EDMX file

If the developer provides an .edmx file (e.g. downloaded from SAP Business Accelerator Hub):

cds import path/to/API_NAME.edmx

This copies the .edmx into srv/external/ and generates a .csn file. Use --as cds if a human-readable .cds file is preferred.

After import, check what was generated in srv/external/:

ls srv/external/

Add the service to package.json under cds.requires if not already done:

"cds"

Use "kind": "odata" for OData v4 services.

Step 2 — Create a consumption view

Create a new .cds file (e.g. apis/<source-system>.cds or srv/external/<name>.cds) with a consumption view — a projection that maps the imported definitions to your domain model, renaming and flattening as needed.

Key conventions:

• Use @federated on entities that will need local data replication (see Step 5).
• Map remote names to your domain names (e.g. BusinessPartner → ID).
• Flatten associations into columns where useful (works over HCQL, not plain OData).
• Add a where clause to pre-filter data if needed.
• If the package annotates the source entity with @cds.minify:'unused-elements', you must override it in your consumption view file:

  annotate S4.A_BusinessPartner with @cds.minify:false;
  Otherwise the projection will have no elements and deployment will fail with "Expecting view to have at least one non-virtual element".

Step 3 — Reference the remote entity from the domain model

In db/schema.cds (or the relevant domain model file), import the consumption view and add the association:

using { sap.capire.xflights as x } from '../apis/capire/xflights';
 
entity Bookings {
  // ... existing elements ...
  Flight : Association to x.Flights;
}

If the remote entity replaces a local entity (e.g. replacing a local Customers with S/4 Business Partner), remove the local entity definition entirely and import the consumption view instead. Then update any existing test data CSVs to use IDs that match the remote mock data (e.g. 000001 instead of 1004155).

Also update any UI annotations (value lists, text arrangements) that reference fields which no longer exist on the new remote-backed entity.

If you also need a back-association from the remote entity to a local one, use an unmanaged association (no foreign keys on the remote side):

extend x.Flights with columns {
  Bookings : Association to many Bookings on Bookings.Flight = $self;
}

Unmanaged associations and virtual/calculated fields are the only safe extensions to remote entities — avoid regular elements or managed associations.

Step 4 — Expose the remote entity in your service (for UIs)

If the remote entity needs to be accessible from a Fiori UI or OData service, expose it as a readonly projection inside your application service:

// In srv/my-service.cds
using { sap.capire.xflights as x } from '../apis/capire/xflights';
 
service MyService {
  // ... existing entities ...
  @readonly entity Flights as projection on x.Flights;
}

Without this, associations to unexposed entities will be cut off.

Step 5 — Wire up the integration handlers (Node.js)

Open (or create) the service handler file (e.g. srv/my-service.js).

5a — Connect to the remote service

The service name passed to cds.connect.to() must exactly match the service name defined in the remote package's index.cds (or the key used in cds.requires for EDMX imports). Always look this up — do not invent an alias.

module.exports = class MyService extends cds.ApplicationService {
  async init() {
    const remote = await cds.connect.to('sap.capire.flights.data')
    // or: cds.connect.to('API_BUSINESS_PARTNER')
    const { Flights, Customers } = this.entities
 
    // ... handlers below ...
 
    return super.init()
  }
}

5b — Delegation (for value helps / on-demand reads)

Delegate incoming READ requests for a remote entity directly to the upstream service. CAP auto-translates queries through the consumption view:

this.on('READ', Customers, req => s4.run(req.query))

5c — Data federation (for joins with local data)

When remote data is needed alongside local data (e.g. displaying flight details alongside bookings), there are two approaches. Ask the user which approach they prefer, explaining the trade-offs:

Option 1: Replicate data locally

Copy remote data into a local persistence table so SQL JOINs work natively.

Pros:

• Full SQL JOIN support (filtering, sorting, paging across local + remote data)
• Fast reads — no network round-trips at query time
• Works offline / when the remote service is unavailable

Cons:

• Data can be stale (only as fresh as the last replication run)
• Requires replication logic and scheduling (initial load + delta sync)
• Consumes local storage
• More moving parts to maintain and monitor

Implementation:

Annotate the consumption view to persist replicated data:

// In db/schema.cds
annotate x.Flights with @cds.persistence.table;

Add replication logic in the service handler (e.g. on startup or on a timer):

const { Flights } = cds.entities('sap.capire.xflights')
let { latest } = await SELECT.one`max(modifiedAt) as latest`.from(Flights)
let touched = await xflights.read(Flights).where`modifiedAt > ${latest || 0}`
if (touched.length) await UPSERT(touched).into(Flights)

Or use @federated (already on the consumption view) together with the CAP Data Federation plugin for automatic replication.

Option 2: Delegate reads to the remote service (no replication)

Handle READ requests for the remote entity by forwarding the query to the upstream service at request time, keeping data always fresh.

Pros:

• Always up-to-date — no stale data
• No local storage or replication logic needed
• Simpler setup for read-only scenarios

Cons:

• No native SQL JOINs across local and remote data
• Every read incurs a network round-trip (latency, availability risk)
• Filtering/sorting across local + remote fields not supported without additional application-level logic
• If the remote service is down, reads fail

Implementation:

Delegate the READ in your service handler:

this.on('READ', Flights, req => xflights.run(req.query))

No @cds.persistence.table annotation is needed. The entity remains virtual from the local DB perspective.

Recommendation

Use replication when the remote data is needed for JOINs, filtering, or sorting together with local data, or when resilience against remote outages matters. Use delegation when data freshness is critical, the dataset is large/volatile, or the entity is only accessed independently (e.g. value helps, detail views).

Important: If the project schema has a local entity with an Association to the remote entity, and that local entity is exposed in a list view (i.e. a SQL JOIN is required), then replication is mandatory — delegation cannot support SQL JOINs across service boundaries. The association will silently return null at runtime. You do not need to ask the user in this case — choose replication.

5d — Outboxed events (reliable fire-and-forget writes)

For write operations to the remote service that don't need a synchronous result:

const xflights_ = cds.outboxed(xflights)
 
this.after('SAVE', Travels, ({ Bookings = [] }) => {
  return Promise.all(Bookings.map(b => {
    let { Flight_ID: flight, Flight_date: date } = b
    return xflights_.send('POST', 'BookingCreated', { flight, date })
  }))
})

This stores the event in a local outbox table within the same DB transaction, then forwards it asynchronously with retries.

Step 6 — Install Cloud SDK prerequisites

Remote HTTP communication requires the SAP Cloud SDK packages. Check if already installed; if not, add them:

npm add @sap-cloud-sdk/connectivity @sap-cloud-sdk/http-client @sap-cloud-sdk/resilience

Step 7 — Verify with cds watch

Start the app and check the log output:

cds watch

Check the connection line carefully. You want to see either:

• mocking <service-name> — CAP is auto-mocking in-process (good for dev)
• connect to <service-name> > odata { url: ... } — CAP is connecting to an external mock process

If you see connect to ... > odata pointing to a URL but no mock is running there, READ requests will fail silently and return empty results. This happens when ~/.cds-services.json has a stale entry from a previous cds mock run. Fix by either:

• Starting the mock: cds mock <service-name>
• Or removing the stale entry so CAP falls back to in-process mocking

Expected: lines like:

[cds] - mocking sap.capire.flights.data { at: [...], ... }

This confirms CAP is auto-mocking the remote service for local development — no upstream service needed during inner-loop development.

To test real delegation (separate mock processes):

# Terminal 1
cds mock apis/capire/xflights.cds
 
# Terminal 2
cds watch

Check that the consuming service now logs connect to ... > hcql (or odata) instead of mocking.

Step 8 — Test queries in cds repl

Quickly validate the integration without starting a UI:

cds repl ./

// Connect to the service
const svc = await cds.connect.to('sap.capire.flights.data')
const { Flights } = cds.entities('sap.capire.xflights')
 
// Read through the consumption view
await svc.read(Flights).limit(3)
 
// Check that column mapping worked (should see domain names, not remote names)

Key patterns summary

Common mistakes to avoid

• Don't expose raw imported entities directly in your service — always go through a consumption view or projection.
• Don't use the view keyword for consumption views — always use entity ... as projection on. (view is a different CDS concept.)
• Don't use @cds.persistence.skip on consumption views — that is a different annotation. Use @federated instead (marks the entity for data federation), and add @cds.persistence.table separately when you want local replication.
• Don't use string entity names in handler registrations — always destructure from this.entities: const { Flights } = this.entities then this.on('READ', Flights, ...) not this.on('READ', 'Flights', ...).
• Don't invent the service name in cds.connect.to() — look it up in the package's index.cds or in cds.requires for EDMX imports.
• Don't modify CQN objects without cloning them first: const q2 = cds.ql.clone(q1).
• Don't add regular elements or managed associations to remote entity extensions — only unmanaged associations, virtual, and calculated fields.
• Don't rely on live cross-service navigation for filtering/sorting without data federation in place — it will fail at runtime.
• Don't forget Cloud SDK packages — without them, remote HTTP calls will error with a missing dependency message.
• Don't project columns that don't exist in the remote mock data — the UI will show blank values. Always cross-check the mock CSV headers against your consumption view columns.
• Don't forget to update test data foreign keys when replacing a local entity with a remote one — old IDs won't match the remote mock data.
• Don't ignore @cds.minify annotations on imported entities — they can strip all elements from your projection, causing deployment failures.