Part III — Capstone Theory (Day‑2, AM)

5.1 Capstone Overview & Success Criteria

Goal: CrewAI agent uses a Langflow tool through the Gateway, with guardrails and observability. Prove: discoverable tool, agent call via gateway, one blocked/one allowed call, RBAC 200/403, logs show correlation/request ID + latency.

5.2 Architecture: Langflow + CrewAI + Gateway

• Langflow: visual builder; flows as HTTP endpoints (/api/v1/run/{flow_id})
• Adapter/Server: bridges Langflow to an MCP tool
• MCP Gateway: registers adapter; enforces policy; exposes catalog
• CrewAI Agent: only calls the Gateway

CrewAI Agent → MCP Gateway → Adapter/Server → Langflow API → Adapter → Gateway → Agent

5.3 Setup & Prerequisites

Gateway on :4444 + JWT; Python 3.11+; Langflow & CrewAI installed.

5.4 Designing the Langflow Tool (I/O and API)

Input: { "text": "..." }
Output: { "summary": "...", "tokens": 123 }

Run Langflow:

pip install langflow
langflow run --host 0.0.0.0 --port 7860
# POST http://localhost:7860/api/v1/run/<flow_id>

5.5 Exposing Langflow as an MCP Tool Server

Adapter must serve /tools (list) and /call/<tool> (invoke + normalize JSON). Register in the Gateway.

5.6 The CrewAI Agent Pattern

Define Agent + Task; call the Gateway (not Langflow) via a small wrapper.

5.7 Hardening

Guardrails (rate limiter + secrets/PII), RBAC (lf.summarize only for analyst), optional OBO, and observability (payload logging, OTEL → Phoenix).