[[{“value”:”
Credit : In a series of super blogs, my colleague @Ashutosh_KSingh has explained how to set up Integration Cell, What is API Centric Integration and How to get started with MCP servers on Integration Suite.
This Blog post builds on this knowledge and works on a scenario comparing SuccessFactors API Consumption via Application vs Agents, powered by MCP servers on Integration Cell.
A New Way to Deliver HR Self-Service
It is 9 AM on a Monday. John Smith has just joined ABC Corp and is sitting through onboarding. The HR presenter mentions the Company Car benefit.
John wants to know what cars he can choose, which vendors are approved, what the lease rates look like, and whether he is already enrolled. The presenter points him to the employee portal.
John opens the portal. It does exactly what it was built to do — he finds a Car Options screen with approved models, vendors, and lease providers, and a second screen showing his enrollment status. The portal works well for the questions it was designed to answer.
Then John asks a follow-up: “Which provider gives me the best deal for a Toyota?” That one is not on any screen. He opens a helpdesk ticket and gets a reply two days later.
This is a familiar and reasonable situation. HR portals are purpose-built, well-governed, and take significant effort to deliver — typically 8 to 12 weeks. They serve the majority of employee needs well. The challenge is the long tail: every new question that was not anticipated at design time either goes unanswered or triggers a new development sprint.
This blog introduces a complementary approach — using Integration Cell, part of SAP Integration Suite’s MCP capability to expose the same SuccessFactors APIs as AI agent tools, so that employees can ask questions in natural language and get answers beyond what any fixed screen provides.
Part 1: The Data Model — SuccessFactors Company Car APIs
Before building anything, it helps to understand what APIs we are working with.
SAP SuccessFactors exposes its benefits domain via OData v2. The Company Car benefit has six entities, arranged in three clean layers.
The Three Layers
┌─────────────────────────────────────────────────────────┐
│ Layer 1 — Reference Data │
│ ┌──────────────────────┐ ┌──────────────────────┐ │
│ │ AllowedModels │ │ RecommendedVendors │ │
│ │ Toyota, Nissan, VW │ │ Hertz, AVIS │ │
│ └──────────────────────┘ └──────────────────────┘ │
│ ┌─────────────────────────────────────┐ │
│ │ LeaseServiceProvider │ │
│ │ AVIS Rental (10.20%), LeasePlan │ │
│ └─────────────────────────────────────┘ │
├─────────────────────────────────────────────────────────┤
│ Layer 2 — Policy │
│ ┌──────────────────────────────────────────────────┐ │
│ │ BenefitCompanyCar │ │
│ │ Lock-in period, maintenance, buyback, EMI freq │ │
│ └──────────────────────────────────────────────────┘ │
├─────────────────────────────────────────────────────────┤
│ Layer 3 — Employee Transactions │
│ ┌──────────────────────┐ ┌──────────────────────┐ │
│ │ CompanyCarEnrollment │ │ CompanyCarClaim │ │
│ │ John's enrollment │ │ John's claims │ │
│ └──────────────────────┘ └──────────────────────┘ │
└─────────────────────────────────────────────────────────┘The full API spec is available on SAP API Business Accelerator Hub: https://hub.sap.com/api/ECGlobalBenefits/overview
Part 2: Building the MCP Server on SAP Integration Suite
SAP Integration Suite’s Integration Cell runtime supports creating MCP servers directly from an OpenAPI specification — no custom code required. Here is how to set it up, step by step.
Step 1 — Get the OpenAPI Spec
Download the API spec from SAP API Business Accelerator Hub (search for “Global Benefits”). The spec is in OpenAPI 2.0 (Swagger) format.
Important: Integration Suite’s MCP server requires OpenAPI 3.0. Use any AI tool (or the Swagger Converter) to upgrade the spec from 2.0 to 3.0 before proceeding.
Step 2 — Create an Integration Package
Log in to your Integration Suite tenant and navigate to Design → Integration Packages. Create a new integration package to hold your MCP server artifact.
Step 3 — Add an MCP Server Artifact
Inside the package, click Add → MCP Server.
In the creation wizard:
- Select HTTP Endpoint with OpenAPI Specification as the connection type
- Set Method to Upload
- Upload your OpenAPI 3.0 spec file
- Set the virtual host path — for example:
<Virtual Host>/sfsf
This is where your MCP endpoint URL will be hosted.
Step 4 — Select Your Tools
After uploading the spec, the wizard discovers all available API paths as potential tools. Search for “Company Car” in the search field and select all 12 related tools. Click Add.
The MCP server artifact is now created.
Step 5 — Configure Tools, Prompts, and Resources
Open the MCP server artifact and switch to the MCP Configuration tab. You will see four sections: Source | Tools | Resources | Prompts.
Under Tools, you can:
- View all selected tools
- Add or remove tools
- Click any tool to inspect its Input Schema and Output Schema
- Edit tool names and descriptions in edit mode
Pro tip: Invest time in editing tool descriptions. The natural language description you write for each tool is what the AI agent reads when deciding which tools to call. A clear description like “Returns the list of car models the company has approved for the benefit program” is far more useful to an agent than “BenefitCompanyCarAllowedModels”.
Similarly, the Prompts and Resources sections can be enriched with additional context per the MCP specification.
Step 6 — Configure Policies
Click the Policies section to view what is applied to the MCP endpoint.
Authentication and Authorization are applied by default.
In the Authentication policy settings, add Basic authentication — this supports Process Integration Runtime (PIR) credentials, which is how agents will authenticate when calling the MCP server. OAuth (client credentials) and Client Certificate authentication are also supported.
For Authorization, keep the default settings. The scope API.invoke is required — the JWT token issued using PIR credentials already contains this scope.
Other policies — traffic management, IP filter — are available from the policy palette. We will cover these in a later article in this series.
Step 7 — Configure the Backend Connection
Before deploying, configure authentication for the SuccessFactors target endpoint.
Click on the HTTP receiver adapter and:
- Set Authentication to Basic
- Enter the Credential Name — for example,
sfsf(a credential stored in the Security Material section of Integration Suite, containing your SuccessFactors username and password)
Step 8 — Deploy
Save the artifact and click Deploy. Select Integration Cell as the Runtime Profile. The virtual host will be pre-populated.
Once deployed, the artifact status changes to Deployed.
Step 9 — Grab the MCP Endpoint URL
Navigate to Monitor → Integrations and APIs, set Runtime to Integration Cell, and click the Manage Integration Content tile → Started.
Your MCP Server artifact appears at the top. The endpoint URL is visible in the right panel — this is the URL you will configure in any AI agent.
Part 3: Connecting an AI Agent
The MCP endpoint can be connected to any MCP-compatible AI client — Joule (SAP’s own AI assistant), Claude Desktop, Cline, GitHub Copilot, and others.
Getting Credentials
The agent authenticates using Process Integration Runtime (PIR) service key credentials. These are the same credentials used to invoke Integration Flows — retrieve them from the service key of your PIR instance on SAP BTP.
Connecting Joule
In this demo we use Joule Work Desktop — SAP’s generative AI assistant and agent orchestrator. Joule supports external MCP server connections.
Once connected, Joule discovers all 12 tools exposed by the MCP server automatically. No further configuration is needed — the agent is ready to reason.
Connecting Claude / Cline
For Claude Desktop or Cline, the MCP configuration looks like this:
{
"mcpServers": {
"sfsf-company-car": {
"type": "http",
"url": "https://<virtual-host>/sfsf",
"headers": {
"Authorization": "Basic <base64-encoded-pir-credentials>"
}
}
}
}
Part 4: The Experience — John Meets His AI Agent
ABC Corp has deployed the MCP server. John opens his AI assistant and types his questions exactly as he would ask a colleague.
Question 1 — “What are my car options?”
“I’m an employee looking to enroll in the company car benefit. What car models can I choose from, which vendors are recommended, and what lease service providers are available?”
The agent reasons silently and calls three tools in parallel: get_allowed_car_models, get_recommended_vendors, get_lease_providers.
Agent responds —
Hi John! Here’s what your company car benefit offers:
Approved Models: Toyota, Nissan, Volkswagen
Approved Vendors: Hertz, AVIS
Lease Financing:
Provider Monthly EMI Rate
AVIS Rental 10.20% LeasePlan Contact HR for rate Installments are paid monthly. Let me know if you’d like details on policy terms!
We can see the conversation with Joule and the tool calls it made on the right side.
Question 2 — “What is my enrollment and claim status?”
“Can you show me the current company car benefit policies, my enrollment details, and whether I have any open claims?”
The agent calls get_car_benefit_policy, get_employee_car_enrollment, and get_employee_car_claims in parallel.
Agent responds:
Policy: Three active car benefit policies, all with monthly installment frequency.
Your Enrollment: You have an active enrollment (ID: 163585), but it is incomplete — your car model, vendor, and EMI amount have not been selected yet.
Claims: No claims on file.
Next step: Complete your enrollment by selecting a model and a financing option. Want help comparing them?
The Follow-up — Where the Agent Extends Beyond the Portal
“Which lease provider gives me the best deal if I go with a Toyota?”
This question sits outside what the portal was designed to show — it would normally mean a helpdesk ticket and a future sprint.
With the MCP agent, no new development is needed. The agent re-reads the lease provider and policy data already retrieved:
The only provider currently showing a configured EMI rate is AVIS Rental at 10.20% — LeasePlan’s rate is not yet set in the system. For a Toyota, AVIS Rental is your best available option. I’d recommend confirming the exact monthly EMI with HR, since it depends on the car’s purchase value.
Zero development. Zero tickets. Answered in seconds.
Part 5: Portal and Agent — Better Together
The portal and the MCP agent are not competing approaches — they serve different needs. The portal delivers a reliable, governed, designed experience for the majority of employee interactions. The agent handles the long tail: the follow-up question, the cross-entity comparison, the query no one thought to build a screen for.
Architecture Comparison
HR Portal:
John opens the portal
│
▼
IT-built portal backend
├── API call 1 → AllowedModels
├── API call 2 → RecommendedVendors
├── API call 3 → LeaseServiceProvider
├── API call 4 → BenefitCompanyCar
├── API call 5 → CompanyCarEnrollment
└── API call 6 → CompanyCarClaim
│
▼
JOIN logic in backend
│
▼
Designed screen renders the data
│
▼
John gets a reliable, consistent answer
for questions the portal was built to answerMCP Agent:
John's natural language question
│
▼
AI Agent (Joule / Claude / Cline)
│
▼
MCP Server on SAP Integration Suite
├── get_allowed_car_models ──→ SF OData API 1
├── get_recommended_vendors ──→ SF OData API 2
├── get_lease_providers ──→ SF OData API 3
├── get_car_benefit_policy ──→ SF OData API 4
├── get_employee_car_enrollment ─→ SF OData API 5
└── get_employee_car_claims ──→ SF OData API 6
│
▼
Agent synthesises, reasons, and responds
│
▼
John gets a conversational, contextual answer
including questions no screen was built forKey Takeaways
-
The MCP server is configured, not coded. SAP Integration Suite’s MCP wizard turns an OpenAPI spec into a deployable MCP endpoint — no custom server code required.
-
Tool descriptions are the interface. The natural language description you write for each tool is what the agent reads when deciding what to call. It replaces the UI specification document.
-
The same security infrastructure you already have applies. PIR credentials, OAuth, and Integration Cell’s policy framework all work with the MCP endpoint out of the box.
-
Any MCP-compatible agent connects. Joule, Claude, Cline, Copilot — once the MCP server is running, the choice of agent is independent of the integration work.
Thanks for reading. Hope you found it useful.
Have questions or want to share how your team is using MCP with SAP? Leave a comment below.
“}]]
│ Layer 1 — Reference Data │
│ ┌──────────────────────┐ ┌──────────────────────┐ │
│ │ AllowedModels │ │ RecommendedVendors │ │
│ │ Toyota, Nissan, VW │ │ Hertz, AVIS │ │
│ └──────────────────────┘ └──────────────────────┘ │
│ ┌─────────────────────────────────────┐ │
│ │ LeaseServiceProvider │ │
│ │ AVIS Rental (10.20%), LeasePlan │ │
│ └─────────────────────────────────────┘ │
├─────────────────────────────────────────────────────────┤
│ Layer 2 — Policy │
│ ┌──────────────────────────────────────────────────┐ │
│ │ BenefitCompanyCar │ │
│ │ Lock-in period, maintenance, buyback, EMI freq │ │
│ └──────────────────────────────────────────────────┘ │
├─────────────────────────────────────────────────────────┤
│ Layer 3 — Employee Transactions │
│ ┌──────────────────────┐ ┌──────────────────────┐ │
│ │ CompanyCarEnrollment │ │ CompanyCarClaim │ │
│ │ John’s enrollment │ │ John’s claims │ │
│ └──────────────────────┘ └──────────────────────┘ │
└─────────────────────────────────────────────────────────┘The full API spec is available on SAP API Business Accelerator Hub: https://hub.sap.com/api/ECGlobalBenefits/overviewPart 2: Building the MCP Server on SAP Integration SuiteSAP Integration Suite’s Integration Cell runtime supports creating MCP servers directly from an OpenAPI specification — no custom code required. Here is how to set it up, step by step.Step 1 — Get the OpenAPI SpecDownload the API spec from SAP API Business Accelerator Hub (search for “Global Benefits”). The spec is in OpenAPI 2.0 (Swagger) format.Important: Integration Suite’s MCP server requires OpenAPI 3.0. Use any AI tool (or the Swagger Converter) to upgrade the spec from 2.0 to 3.0 before proceeding.Step 2 — Create an Integration PackageLog in to your Integration Suite tenant and navigate to Design → Integration Packages. Create a new integration package to hold your MCP server artifact.Step 3 — Add an MCP Server ArtifactInside the package, click Add → MCP Server.In the creation wizard:Select HTTP Endpoint with OpenAPI Specification as the connection typeSet Method to UploadUpload your OpenAPI 3.0 spec fileSet the virtual host path — for example: <Virtual Host>/sfsf This is where your MCP endpoint URL will be hosted.Step 4 — Select Your ToolsAfter uploading the spec, the wizard discovers all available API paths as potential tools. Search for “Company Car” in the search field and select all 12 related tools. Click Add. The MCP server artifact is now created.Step 5 — Configure Tools, Prompts, and ResourcesOpen the MCP server artifact and switch to the MCP Configuration tab. You will see four sections: Source | Tools | Resources | Prompts. Under Tools, you can:View all selected toolsAdd or remove toolsClick any tool to inspect its Input Schema and Output SchemaEdit tool names and descriptions in edit modePro tip: Invest time in editing tool descriptions. The natural language description you write for each tool is what the AI agent reads when deciding which tools to call. A clear description like “Returns the list of car models the company has approved for the benefit program” is far more useful to an agent than “BenefitCompanyCarAllowedModels”.Similarly, the Prompts and Resources sections can be enriched with additional context per the MCP specification.Step 6 — Configure PoliciesClick the Policies section to view what is applied to the MCP endpoint. Authentication and Authorization are applied by default.In the Authentication policy settings, add Basic authentication — this supports Process Integration Runtime (PIR) credentials, which is how agents will authenticate when calling the MCP server. OAuth (client credentials) and Client Certificate authentication are also supported. For Authorization, keep the default settings. The scope API.invoke is required — the JWT token issued using PIR credentials already contains this scope.Other policies — traffic management, IP filter — are available from the policy palette. We will cover these in a later article in this series. Step 7 — Configure the Backend ConnectionBefore deploying, configure authentication for the SuccessFactors target endpoint.Click on the HTTP receiver adapter and:Set Authentication to BasicEnter the Credential Name — for example, sfsf (a credential stored in the Security Material section of Integration Suite, containing your SuccessFactors username and password)Step 8 — DeploySave the artifact and click Deploy. Select Integration Cell as the Runtime Profile. The virtual host will be pre-populated. Once deployed, the artifact status changes to Deployed. Step 9 — Grab the MCP Endpoint URLNavigate to Monitor → Integrations and APIs, set Runtime to Integration Cell, and click the Manage Integration Content tile → Started.Your MCP Server artifact appears at the top. The endpoint URL is visible in the right panel — this is the URL you will configure in any AI agent. Part 3: Connecting an AI AgentThe MCP endpoint can be connected to any MCP-compatible AI client — Joule (SAP’s own AI assistant), Claude Desktop, Cline, GitHub Copilot, and others.Getting CredentialsThe agent authenticates using Process Integration Runtime (PIR) service key credentials. These are the same credentials used to invoke Integration Flows — retrieve them from the service key of your PIR instance on SAP BTP. Connecting JouleIn this demo we use Joule Work Desktop — SAP’s generative AI assistant and agent orchestrator. Joule supports external MCP server connections. Once connected, Joule discovers all 12 tools exposed by the MCP server automatically. No further configuration is needed — the agent is ready to reason. Connecting Claude / ClineFor Claude Desktop or Cline, the MCP configuration looks like this:{
“mcpServers”: {
“sfsf-company-car”: {
“type”: “http”,
“url”: “https://<virtual-host>/sfsf”,
“headers”: {
“Authorization”: “Basic <base64-encoded-pir-credentials>”
}
}
}
}
Part 4: The Experience — John Meets His AI AgentABC Corp has deployed the MCP server. John opens his AI assistant and types his questions exactly as he would ask a colleague.Question 1 — “What are my car options?””I’m an employee looking to enroll in the company car benefit. What car models can I choose from, which vendors are recommended, and what lease service providers are available?”The agent reasons silently and calls three tools in parallel: get_allowed_car_models, get_recommended_vendors, get_lease_providers.Agent responds — Hi John! Here’s what your company car benefit offers:Approved Models: Toyota, Nissan, VolkswagenApproved Vendors: Hertz, AVISLease Financing:Provider Monthly EMI Rate AVIS Rental10.20%LeasePlanContact HR for rateInstallments are paid monthly. Let me know if you’d like details on policy terms! We can see the conversation with Joule and the tool calls it made on the right side. Question 2 — “What is my enrollment and claim status?” “Can you show me the current company car benefit policies, my enrollment details, and whether I have any open claims?”The agent calls get_car_benefit_policy, get_employee_car_enrollment, and get_employee_car_claims in parallel.Agent responds:Policy: Three active car benefit policies, all with monthly installment frequency.Your Enrollment: You have an active enrollment (ID: 163585), but it is incomplete — your car model, vendor, and EMI amount have not been selected yet.Claims: No claims on file.Next step: Complete your enrollment by selecting a model and a financing option. Want help comparing them? The Follow-up — Where the Agent Extends Beyond the Portal”Which lease provider gives me the best deal if I go with a Toyota?”This question sits outside what the portal was designed to show — it would normally mean a helpdesk ticket and a future sprint.With the MCP agent, no new development is needed. The agent re-reads the lease provider and policy data already retrieved:The only provider currently showing a configured EMI rate is AVIS Rental at 10.20% — LeasePlan’s rate is not yet set in the system. For a Toyota, AVIS Rental is your best available option. I’d recommend confirming the exact monthly EMI with HR, since it depends on the car’s purchase value.Zero development. Zero tickets. Answered in seconds.Part 5: Portal and Agent — Better TogetherThe portal and the MCP agent are not competing approaches — they serve different needs. The portal delivers a reliable, governed, designed experience for the majority of employee interactions. The agent handles the long tail: the follow-up question, the cross-entity comparison, the query no one thought to build a screen for.Architecture ComparisonHR Portal:John opens the portal
│
▼
IT-built portal backend
├── API call 1 → AllowedModels
├── API call 2 → RecommendedVendors
├── API call 3 → LeaseServiceProvider
├── API call 4 → BenefitCompanyCar
├── API call 5 → CompanyCarEnrollment
└── API call 6 → CompanyCarClaim
│
▼
JOIN logic in backend
│
▼
Designed screen renders the data
│
▼
John gets a reliable, consistent answer
for questions the portal was built to answerMCP Agent:John’s natural language question
│
▼
AI Agent (Joule / Claude / Cline)
│
▼
MCP Server on SAP Integration Suite
├── get_allowed_car_models ──→ SF OData API 1
├── get_recommended_vendors ──→ SF OData API 2
├── get_lease_providers ──→ SF OData API 3
├── get_car_benefit_policy ──→ SF OData API 4
├── get_employee_car_enrollment ─→ SF OData API 5
└── get_employee_car_claims ──→ SF OData API 6
│
▼
Agent synthesises, reasons, and responds
│
▼
John gets a conversational, contextual answer
including questions no screen was built forKey TakeawaysThe MCP server is configured, not coded. SAP Integration Suite’s MCP wizard turns an OpenAPI spec into a deployable MCP endpoint — no custom server code required.Tool descriptions are the interface. The natural language description you write for each tool is what the agent reads when deciding what to call. It replaces the UI specification document.The same security infrastructure you already have applies. PIR credentials, OAuth, and Integration Cell’s policy framework all work with the MCP endpoint out of the box.Any MCP-compatible agent connects. Joule, Claude, Cline, Copilot — once the MCP server is running, the choice of agent is independent of the integration work.Thanks for reading. Hope you found it useful.Have questions or want to share how your team is using MCP with SAP? Leave a comment below.”}]] Read More Technology Blog Posts by SAP articles
#SAPCHANNEL