Fred

A Sample Demo Architecture

Context

This page is a documentation sample derived from a live demo session.

It describes a realistic integration pattern for Fred around a postal operations use case. Even if the original demo was custom, the architecture principles are reusable.

Overview

This demo is based on a La Poste agent running in Fred Agentic Backend (custom TrackingAgent or BasicReAct) that orchestrates the conversation, calls business tools through MCP, and can enrich answers with content from Knowledge Flow (private demo documents).

The goal is to demonstrate an agent able to:

  • understand customer requests (tracking, incident, resolution action),
  • gather reliable operational context,
  • propose concrete actions (rerouting, rescheduling, and so on),
  • request human validation (HITL),
  • and return rich responses (text, map, options).

The two MCP servers used in this demo

The demo relies on two complementary MCP servers, each with a clearly separated business role.

1. Parcel Operations & Customer Service MCP

This server covers postal business actions: parcel tracking, pickup-point lookup, rerouting to pickup point, delivery rescheduling, customer notification, compensation estimation, claim creation, and more.

In practice, this is what enables the agent to propose and execute resolution actions for delayed parcels or incidents.

2. Telemetry & Logistics Simulation MCP

This server provides real-time simulated operational context: tracking snapshots, tracking events, hub status, vehicle position, route geometry, locker occupancy, and simulation progress.

In practice, this enables the agent to reason on the current logistics state (hub congestion, waiting vehicle, route progression, and so on) and feed map views in the demo.

Why two MCP servers instead of one?

The split is intentional:

  • one MCP for business capabilities and customer actions,
  • one MCP for operational context and telemetry.

This architecture makes the demo easier to understand and better reflects real environments, where transactional business systems and operational tracking systems are often separated.

Simplified View

%%{init: {"themeVariables": {"fontSize": "16px"}}}%%
flowchart TB

  subgraph UI["🧑 UI"]
    direction TB
    UChat["Demo chat"]
    UAdmin["Agent Hub
(config + agent graph)"]
  end

  subgraph AG["🧠 Fred Agentic Backend"]
    direction TB
    Api["API Chat / WebSocket"]
    Fred["Fred Orchestrator"]
    Agent["La Poste Agent
(TrackingAgent or BasicReAct)"]
  end

  subgraph EXT["Business and Knowledge Services"]
    direction LR

    subgraph MCP["🔌 MCP La Poste"]
      direction TB
      MCPTrack["Parcel Tracking and Incidents"]
      MCPOps["Delivery Actions and Customer Service"]
    end

    subgraph KF["📚 Knowledge Flow"]
      direction TB
      KFApi["API RAG"]
      Corpus["Private corpus
(procedures, pricing, scripts)"]
    end
  end

  UChat --> Api
  UAdmin --> Api

  Api --> Fred
  Fred --> Agent

  Agent --> MCPTrack
  Agent --> MCPOps
  Agent --> KFApi
  KFApi --> Corpus

  style UI fill:#EAF2FF,stroke:#3B4A6B,stroke-width:1.2px
  style AG fill:#FFEFE0,stroke:#7A4A21,stroke-width:1.2px
  style MCP fill:#FFF7E6,stroke:#7A4A21,stroke-width:1.2px
  style KF fill:#E9FFE9,stroke:#2E6B2E,stroke-width:1.2px

Full Architecture View

%%{init: {"themeVariables": {"fontSize": "15px"}}}%%
flowchart TB

  %% =========================
  %% IAM (top layer)
  %% =========================
  subgraph SEC["🔐 Identity & Access"]
    KC["Keycloak
(OIDC / JWT / SSO)"]
    FGA["OpenFGA
(fine-grained authorization / RBAC/ABAC)"]
  end

  %% =========================
  %% Frontend
  %% =========================
  subgraph UI["🧑 UI"]
    Chat["Chat + HITL cards + GeoMap"]
    Hub["Agent Hub
(edit + graph)"]
    KfPages["Knowledge Flow UI
(corpus management)"]
  end

  %% =========================
  %% Agentic Platform
  %% =========================
  subgraph AG["🧠 Agentic Platform (Fred)"]
    AGApi["API Chat / WS / Agents"]
    Orch["Fred Orchestrator"]
    PostalAgent["La Poste Agent
(custom / ReAct)"]
    Temporal["Temporal
(long-running / async workflows)"]
    AGPg["PostgreSQL
(sessions, history, config)"]
  end

  %% =========================
  %% Business services
  %% =========================
  subgraph MCP["🔌 Business MCP Services"]
    Mcp1["Parcel Tracking and Incidents"]
    Mcp2["Delivery Actions / Customer Service"]
  end

  subgraph KF["📚 Knowledge Flow"]
    KFApi["RAG / Ingestion / Search API"]
  end

  %% =========================
  %% Storage / Data layer
  %% =========================
  subgraph DATA["💾 Data & Storage"]
    KFPg["PostgreSQL
(metadata)"]
    Vec["Vector Store
(embeddings / similarity)"]
    S3["Object Store / S3
(documents)"]
  end

  %% =========================
  %% Flows
  %% =========================

  %% Auth
  Chat --> KC
  Hub --> KC
  KfPages --> KC

  KC --> AGApi
  KC --> KFApi

  AGApi --> FGA
  KFApi --> FGA

  %% Frontend to backend
  Chat --> AGApi
  Hub --> AGApi
  KfPages --> KFApi

  %% Agentic internals
  AGApi --> Orch
  Orch --> PostalAgent
  AGApi --> AGPg
  Orch --> Temporal

  %% Tools / Knowledge
  PostalAgent --> Mcp1
  PostalAgent --> Mcp2
  PostalAgent --> KFApi

  %% Knowledge storage
  KFApi --> KFPg
  KFApi --> Vec
  KFApi --> S3

  %% Styles
  style UI fill:#EAF2FF,stroke:#3B4A6B,stroke-width:1.2px
  style SEC fill:#F5F5F5,stroke:#666,stroke-width:1.2px
  style AG fill:#FFEFE0,stroke:#7A4A21,stroke-width:1.2px
  style MCP fill:#FFF7E6,stroke:#7A4A21,stroke-width:1.2px
  style KF fill:#E9FFE9,stroke:#2E6B2E,stroke-width:1.2px
  style DATA fill:#F9F9F9,stroke:#777,stroke-width:1.2px

  style PostalAgent fill:#FFF2CC,stroke:#6B5A1F,stroke-width:1.4px
  style Orch fill:#FFD9B3,stroke:#7A4A21,stroke-width:1.4px
Prev
Roadmap
Next
Add your agent