In the previous post I built btp-agent: a single LangGraph agent in a CAP app that wires up five BTP MCP servers (Cloud Integration, AI Core, BTP Core, Audit Log, Cloud Foundry V3) and lets users chat with their full BTP landscape. It worked, but it hit some limitations. With all five MCP servers plugged in, the model saw well over a hundred tools on every turn. Latency climbed, the model reached number of tool limits and the single orchestrator prompt got asked to be an expert in integration flows, AI deployments, audit events and CF applications all at once. For a live demo that is survivable. For anything that wants to scale, it isn’t.
So I took the agent apart.
In the previous post I built btp-agent: a single LangGraph agent in a CAP app that wires up five BTP MCP servers (Cloud Integration, AI Core, BTP Core, Audit Log, Cloud Foundry V3) and lets users chat with their full BTP landscape. It worked, but it hit some limitations. With all five MCP servers plugged in, the model saw well over a hundred tools on every turn. Latency climbed, the model reached number of tool limits and the single orchestrator prompt got asked to be an expert in integration flows, AI deployments, audit events and CF applications all at once. For a live demo that is survivable. For anything that wants to scale, it isn’t.So I took the agent apart. Read More Technology Blog Posts by Members articles
#SAP
#SAPTechnologyblog
