Chapter 6 - Implementing the Plan
This is the chapter you have been building toward. Every concept covered so far — prompting, MCP connectivity, agentic workflows, CLI skills, and structured planning — comes together here. You will take the plan you designed in Chapter 5 and execute it against a real SAP system, letting the AI agent do the heavy lifting while you stay in control through targeted review and validation.
- Reading time: 10 minutes
- Practical time: 90 minutes
Reading
Read this material before starting the hands-on part:
Briefs
| Priority | Read This | Why It Matters |
|---|---|---|
| 1 | Tools Comparison: MCP vs Skill vs Agent vs Agent Context File | Clarifies which layer of your setup owns each responsibility — especially relevant as the agent now makes live changes through MCP. |
Reading Outcomes
After the reading, you should be able to:
- Explain which tool layer (MCP, Skill, Agent, Agent Context File) is responsible for each action during plan execution.
- Describe why restricting the MCP scope to a specific package and transport is important before running an agentic implementation.
Practical
IMPORTANT
Prerequisite: You must have completed Chapter 5 and have a validated implementation plan before starting this chapter.
Exercise 1 - Create a Package and Transport Request, and Restrict MCP Scope
Before the agent makes any changes to the SAP system, you need a dedicated package and transport request to contain all generated objects, and you need to configure your MCP server to write only within that scope.
Step 1 - Create the package
- Open the SAP system (e.g. via Eclipse or SAP GUI).
- Create a new package using your initials as a prefix, for example
Z<YOUR_INITIALS>_AILAB. - Assign the package to a software component appropriate for the system (e.g.
HOMEfor local development or a project-specific component).
Step 2 - Create a transport request
- In the package creation dialog (or via transaction
SE09/SE10), create a new Workbench transport request. - Give it a meaningful description, for example:
AI Lab - Chapter 6 implementation. - Note down the transport request number (e.g.
<SID>K9XXXXX) — you will need it in the next step.
Step 3 - Restrict MCP scope
Update your MCP server configuration to limit all write operations to the package and transport you just created. Add the following environment variables to your arc1 server entry:
"SAP_ALLOWED_PACKAGES": "Z<YOUR_INITIALS>_AILAB",
"SAP_ALLOWED_TRANSPORTS": "<YOUR_TRANSPORT_REQUEST_NUMBER>"- Save the configuration and reload the MCP server so the new environment variables take effect.
- Verify the scope restriction is active by asking Copilot which package and transport it will use for write operations.
Note
Restricting the package and transport ensures that every object created by the agent lands in a controlled, reversible scope. If anything goes wrong, you can release or delete the transport request without affecting other work.
Exercise 2 - Implement the Plan from Chapter 5
With your package, transport, and restricted MCP scope in place, it is time to execute the plan you designed in Chapter 5.
Goal
Deliver the working SAP feature described in your Chapter 5 plan, implemented step by step by the AI agent under your supervision.
Steps
- Open Copilot CLI (agent mode) and attach your Chapter 5 plan as context.
- Use the following prompt to start the implementation:
I have a validated implementation plan. Please execute it step by step.
For each step:
- Use the ARC-1 MCP server for all SAP read and write operations.
- Write all new objects to package Z<YOUR_INITIALS>_AILAB and transport <YOUR_TRANSPORT_REQUEST_NUMBER>.
- After completing each step, briefly confirm what was done and what comes next.
- Pause and ask for my approval before any destructive or irreversible operation.- Review each agent action as it is executed. Approve MCP tool calls when prompted.
- After each major step is completed, verify the output in the SAP system (via ADT or SAP GUI) before proceeding.
- Once all steps are complete, confirm the transport request contains all expected objects.
Important
You remain the reviewer throughout. The agent proposes and executes — you validate. Do not approve MCP write operations without checking that the proposed change matches your plan.
Checkpoint
The exercise is complete when:
- All objects defined in the Chapter 5 plan exist in the SAP system under your package.
- All objects are assigned to your transport request.
- The implemented feature behaves as specified in the plan.
What's Next
Congratulations — you have completed the SAP AI Coding Lab. You have gone from setting up a secure SAP-connected environment to planning and shipping a real feature with AI assistance. The skills you practised here — scoped MCP configuration, structured planning, agentic execution, and supervised validation — are directly applicable to production SAP development workflows.