AnQiCMS MCP Server实战：Let Claude and Cursor directly operate your CMS

To allow Claude and Cursor, etc., AI clients to directly operate AnQiCMS through MCP Server, it is necessary to complete the authorization configuration and understand the structure of the intent directory. AnQi Content Management System (AnQiCMS) is built-in with MCP Server, providing a standardized capability exposure channel for AI integration

The architecture of MCP Server

MCP Server is the standard interface layer exposed to the outside by AnQiCMS, allowing external AI clients to call system functions via a unified protocol. The tool layer is unified and convergent based on the intent directory, integrating the capabilities scattered in content management, SEO settings, site operation and maintenance modules into a structured tool list.

The intent directory organizes tools by functional domain, such as content creation, classification management, tag operation, message handling, file upload, and so on. Each tool defines clear input parameters and return formats. The AI client does not need to understand the details of the underlying API. It only needs to initiate calls according to the intent name and parameter descriptions. This design reduces the threshold for AI integration. Developers only need to configure MCP connection information, and the AI can automatically discover available tools.

Authorization configuration steps

The MCP interface involves read and write operations on CMS data and must undergo strict authentication. AnQiCMS uses an authorization mechanism to ensure that only authorized AI clients can access the system.

The configuration steps are as follows:

1. Generate access token: Create a new access token on the AnQiCMS background API key management page, record the generated token string. The token should be kept safely and should not be written in plaintext to the code repository or log files.

2. Configure MCP clientIn Claude Desktop, Cursor, or any other AI client that supports MCP, add the MCP server address and access credentials for AnQiCMS. The client will automatically carry the authentication information in the HTTP Header with each request.

3. Test connectionUse the AI client to send a simple query command to verify that the MCP connection is normal. A successful response should return structured data instead of error information.

The permission scope of the certificate should match the usage scenario. For read-only scenarios such as content retrieval, it is possible to configure only open query tools; for write operation scenarios such as batch publishing, additional permissions for content creation and update must be enabled.

Round-based batch approval mechanism

The risk of write operations is higher than that of read operations. AnQiCMS's MCP Server supports round-based batch approval mechanism. When the AI client sends multiple write operation requests consecutively, the system will merge these operations into one approval round. The administrator only needs to confirm once to approve all changes in the batch.

The configuration options for round-level approval include automatic approval thresholds, high-risk operation blocking, and approval timeout strategies, etc. High-risk operations such as deleting articles, modifying system settings, etc., regardless of the number, require separate approval.

Example of external AI client connection

The typical process of using Claude Desktop to connect to the AnQiCMS MCP Server: Add MCP server configuration in settings, enter the service address and access credentials; after restarting, you should see the AnQiCMS exposed intent tools in the tool list; issue commands to Claude like 'help me create an article about SEO optimization'; Claude will call the content creation tool of MCP and submit it to AnQiCMS; the system will return the creation result and provide a preview link.

The access method of Cursor code editors is similar, but Cursor focuses more on code-level operations, suitable for tasks such as template development, API debugging, and other technical tasks.

Permission control for high-risk domain operations

The MCP interface sets the exposure range according to the intent domain, and high-risk domains such as backup, upgrade, multi-site management, and other full-site operations are turned off by default. This default off design follows the principle of least privilege to avoid the AI executing dangerous operations due to misunderstanding user instructions.

Typical high-risk operations include data backup and recovery, system upgrade, multi-site management, and user permission adjustment. It is recommended to perform these operations outside of working hours and to notify relevant team members in advance.

Expansion capabilities of the AI skill system

In addition to MCP Server, AnQiCMS also built-in AI skill system (Skills), covering content planning, SEO analysis, batch operations, site health check, troubleshooting, template development, and API development. Skills are expandable predefined workflows, and AI Agent can independently orchestrate the call of multiple skills to complete complex tasks.

Built-in AI Chat intelligent dialogue as a unified interaction entry, users can describe their needs through natural language, and AI will automatically select the appropriate skills and MCP tools to complete the task. This layered architecture allows ordinary users to enjoy the efficiency improvement brought by AI.

Common questions

Q: What is the difference between MCP Server and local API?

A: MCP Server is an interface layer based on standard protocols, designed specifically for AI clients, providing semantic tool descriptions and batch approval mechanisms. Local API is more suitable for traditional application integration and requires developers to manually handle authentication, parameter verification, and error retries.

Q: Will there be conflicts if multiple AI clients access simultaneously?

A: No. The MCP Server handles requests sequentially, and concurrent requests will be queued for execution. The JWT authentication mechanism ensures that each request has an independent identity identifier, and the operation log can trace the source of each change.

Q: How to undo the incorrect operation executed by AI?

A: AnQiCMS's document recycling bin feature retains deleted content, and administrators can restore it within a certain period of time. For modification operations, you can locate the change point through the operation log and use the site-wide replacement tool to batch correct. It is recommended to create a data backup before executing batch write operations.