MCP-BFF Service

Scope

This page documents MCP-BFF behavior (AI chat + tool orchestration layer).
It is different from the classical DataView-oriented BFF service.

For DataView/CQRS read-model design, see BFF Service.

---

What MCP-BFF Does

MCP-BFF acts as a unified interface for AI-driven product workflows:

• chat message processing
• tool discovery and routing across services
• docs retrieval endpoints
• AI completion helpers
• service status and diagnostics

---

Endpoint Families (Conceptual)

• Chat: message and stream flows, conversation lifecycle
• Tools: list, service-scoped list, call, refresh, status
• Docs: service/API documentation retrieval
• AI Complete: assisted request/body generation helpers
• Health/Status: service readiness and integration diagnostics

Use generated API docs for exact route signatures in your environment.

---

Tenant-Aware Behavior

In multi-tenant projects, MCP-BFF propagates tenant context in routed requests.

Design implications:

• always send tenant context from frontend session/routing layer
• keep tool calls tenant-safe by default
• verify cross-service calls preserve tenant headers/metadata

---

Practical Usage Guidance

• Use MCP-BFF for orchestrated AI interactions, not direct domain CRUD replacement.
• Keep sensitive APIs/tool calls role-protected.
• Refresh tool registry after major service generation updates.
• Prefer structured prompts that reference concrete service/API names.

---

Operational Notes

• Validate health/status endpoints during startup and after deployments.
• Check logs/diagnostics endpoints when tool discovery appears incomplete.
• Treat streaming and non-streaming chat paths as separate client concerns.

---

Related Pages

• BFF Service
• Consuming Your Backend
• Machine-to-Machine Communication