Kingdee MCP bridge for AI assistants and K/3 Cloud workflows
kingdee-k3cloud-mcp by Adamzhang1987 is an MCP server that connects AI assistants to Kingdee K/3 Cloud ERP for conversational access and automation. It exposes Kingdee's Web API as standardized MCP tools so language models can query inventory, process sales orders, and run data analysis via natural language. The tool bundles 15 specialized MCP tools, universal form_id handling, pagination and streaming primitives, session recovery, multiple transports, and a read-only toggle. It targets enterprise developers, data analysts, and IT administrators who need AI-driven ERP interactions without bespoke API coding.
It converts Kingdee into a set of MCP-accessible tools for AI agents
The tool exposes Kingdee Web API endpoints as MCP tools, letting models perform common ERP tasks through single-parameter form calls. The distribution includes 15 specialized MCP tools covering operations such as query, add, submit, audit, and delete. Example use cases range from simple inventory checks to processing sales documents and running ad-hoc data queries, all by addressing Kingdee forms via a universal form_id parameter.
Large-data queries and long operations are handled with primitives for AI contexts
Built-in pagination and streaming reduce the burden on model context windows. The server provides automatic pagination, date slicing, and a query_bill_to_file primitive to stream results to disk, which suits high-volume exports. Automatic session recovery is present to maintain continuity for long-running workflows, so agents can resume operations without manual session re-establishment.
Installation and inputs require a developer-controlled environment
Deployment needs a Python 3.10+ environment and the uv package manager. Integration requires valid Kingdee K/3 Cloud Web API credentials (URL, Account ID, Username, App ID, App Secret) and correct form_id values to target specific modules. Transport options include stdio for local use, and SSE or streamable-http for remote sharing, so network configuration and credential management affect successful operation.
Fits MCP client workflows and has visible community signals
The server integrates with MCP-compliant clients such as Claude Desktop, Cursor, and Claude Code, and it pairs with kingdee-k3cloud-skill for ERP field and workflow knowledge. Its listing in major MCP server directories and active updates on GitHub indicate ongoing maintenance and niche adoption. Administrators should treat remote transports and write-capable modes with appropriate access controls during integration.
A practical, developer-focused bridge for AI-enabled Kingdee workflows
The tool is a maintained option for teams building MCP-based agents that need conversational access to Kingdee, supported by directory listings and active GitHub updates. It suits groups with developer resources to manage deployment and security. Expect to validate any agent outputs and to enforce internal controls before using write-capable modes in production environments.
