HARQIS Hands-On: Complete Practice of Automated Workflows, MCP Integration, and Claude AI Agents

HARQIS-work is an open-source self-hosted AI automation portfolio by brianbartilet, focusing on real-world application scenarios. It integrates Claude AI, MCP protocol, and automated workflows, providing developers with a complete practical reference for AI application implementation. The project covers multi-domain solutions such as intelligent document processing, code review, and intelligent customer service, and is a directly deployable practical system.

With the maturity of large language model capabilities, AI agents and automated workflows are moving from proof-of-concept to practical applications. The harqis-work project (standing for "Human-Assisted Real-time Query and Intelligence System") demonstrates how to integrate Claude AI, MCP protocol, and automation tools into actual work processes. Unlike demo projects, it is a complete deployable solution covering automation from personal productivity to team collaboration, providing developers with implementation references.

MCP Protocol: Standard Interface Between AI and Tools

MCP is an open protocol launched by Anthropic, standardizing the interaction between AI and external tools/data sources. In the project, it acts as a bridge between AI agents and services, supporting modes such as tool calling, resource access, and prompt templates, simplifying the integration of new tools.

Claude AI Agent Architecture

A multi-level agent architecture is built: the base layer encapsulates the Claude API, the agent layer supports task planning/execution/monitoring and single/multi-agent collaboration, and also implements persistent long-term memory, semantic retrieval, and context integration.

Automated Workflow Engine

A built-in lightweight engine supports declarative syntax to define complex processes, responds to triggers like scheduled tasks and file changes, and has features such as parallel tasks, conditional branches, and exception recovery.

Intelligent Document Processing

Automatically triggered when a new document is added: OCR text extraction → Claude analysis and summarization → classification and archiving into the knowledge base, realizing collaboration between AI and traditional tools.

Code Review and Refactoring

When a Pull Request is created, the agent automatically obtains changes, performs static analysis, generates review comments, and even generates fix code and integrates with IDEs.

Intelligent Customer Service and Tickets

Receives multi-channel inquiries: Claude understands intent → automatically retrieves knowledge base for replies → transfers to humans when necessary, maintaining cross-channel conversation history.

Data Pipelines and Reports

Regularly extracts multi-source data for cleaning and transformation → Claude generates natural language reports, highlighting key insights.

Modular Service Architecture

Adopts a microservices architecture; core services include API gateway, workflow engine, agent manager, etc., which can be deployed and expanded independently.

Data Security and Privacy

The self-hosted solution prioritizes security: end-to-end encryption, fine-grained access control, audit logs, and sensitive data transmission and storage are protected.

Observability and Operations

Built-in structured logs, performance metrics, distributed tracing, and alerts; monitors workflow and agent status via dashboards.

Custom Tool Integration

Provides extension interfaces; any API service can be integrated by implementing the MCP protocol, including detailed guides and sample code.

Workflow Templates and Best Practices

Maintains a library of templates for common scenarios, which can be used directly or customized; summarizes reliable workflow design best practices (error handling, idempotency, timeout management).

Comparison with Similar Projects

• Compared to LangChain/AutoGen: More practical, it is a directly deployable solution rather than a general framework, suitable for teams with clear automation needs.
• Compared to commercial platforms (Zapier/Make): Higher flexibility and data control, self-hosted with no vendor lock-in.

Summary and Outlook

harqis-work provides a practical reference for AI automation, demonstrating how to integrate technologies into real scenarios to create value. We look forward to more such projects to drive AI from the lab to the production environment.