Admin AI & Automation

Finqu provides an open, extensible architecture for commerce automation built on the Model Context Protocol (MCP). Rather than locking automation inside proprietary silos, Finqu gives developers two core integration capabilities:

Drive Finqu with any agent (Outbound via Admin MCP): Connect external AI agents, IDE assistants (Cursor, Codex, Claude Desktop), or custom agent applications to operate Finqu admin tools over Streamable HTTP.
Supercharge Finqu AI with any tool (Inbound via MCP Tools): Connect your external APIs, ERPs, CRMs, and internal microservices into Finqu AI so that native assistants and automated workers can query and mutate outside systems.

Between these connected tools and your business instructions lies Finqu’s automation engine in Work Hub: Workers, Playbooks, and Events.

Two-Way MCP Architecture

Core Capabilities

Capability	Integration Vector	What it enables
Admin MCP	Outbound	Connect Claude, Cursor, Codex, or custom agents to operate store catalog, orders, themes, and settings
Connecting Agents	Outbound	Ready-to-run configurations and SDK recipes in TypeScript and Python
Inbound MCP Tools	Inbound	Register external MCP servers to expose your ERP, warehouse, or CRM tools to Finqu agents
Authoring Workers	Automation	Build specialized background agents: Analyst (data reports), Planner (action plans), and Executor (mutations)
Authoring Playbooks	Automation	Define multi-step pipelines with variable passing, tool whitelists, and human-in-the-loop approval gates
Events	Automation	Trigger workers and playbooks automatically when store actions occur or on cron schedules
Success Agent	Interactive	Merchant-facing conversational assistant equipped with store context and connected MCP tools
Guidelines	Governance	Define shared operational rules, margin floors, and brand voice that govern all agent behavior

The Automation Engine

Inside Finqu admin, Work Hub provides the runtime that connects your tools, agents, and instructions:

Success Agent: The interactive chat assistant in the admin. It answers merchant questions, diagnoses store health, and executes approved tasks using both native and inbound MCP tools.
Workers: Focused, task-oriented background agents. Each worker specializes in a role (Analyst observes, Planner strategizes, Executor mutates) with an explicit tool whitelist.
Playbooks: Structured, deterministic workflows. Playbooks define sequential steps performed by workers, pass outputs between steps, and support human confirmation gates for safety.
Events: The trigger mechanism. Configure events to run workers or playbooks automatically when orders are placed, stock runs low, or on recurring cron schedules.

Developer Quick Start

Depending on your goal, start with the relevant guide:

Connect your external agent to Finqu: Follow Connecting Agents to configure Claude Desktop, Cursor, or an SDK harness. Review Best Practices for channel resolution and confirmation gates.
Connect your external system into Finqu AI: Follow Inbound MCP Tools to register your MCP server and make external tools available to Finqu.
Automate recurring operations: Read Authoring Workers to define task-specific agents, then chain them with Authoring Playbooks and trigger them via Events.

Shopping Agents

For customer-facing shopping assistants that browse products, manage carts, and drive checkout, see Agentic Commerce. Storefront agent interactions are handled through the Universal Commerce Protocol (UCP), separate from Admin MCP.