Feature preview
The intent directory manages the callable operation surface: interfaces converge according to the intent domain, and external clients cannot read unregistered capabilities; round-level approval manages the actual write operations that occur in a round, merging multiple changes into one confirmation to avoid误伤 caused by approving each change individually. The high-risk areas at the station level are closed by default and need to be explicitly opened. It is a safe practice to keep multi-site and backup operations closed in the initial stage of integration.
The content management system accesses external AI clients such as Claude, Cursor, etc. through MCP, what operations do the intent directory and round-level approval manage separately?In short, the intent directory controls what operations the client can see and call, and the round-level approval controls whether the changes to be implemented in a round are allowed.The former is range control, the latter is execution control, the problems solved by the two gates are different, and the sequence is also first range and then execution. AnQiCMS (AnQi CMS) built-in MCP Server, the tool surface is unified convergence according to the intent directory, authorization goes through Bearer method and supports round-level operation approval, write operations will be merged and confirmed in one request.
The access method without convergence usually distributes all interfaces to the client, and the caller decides whether to use them. The approach of the intent directory is the opposite: only the operations registered in the directory will be exposed, and the client receives a list of capabilities divided by intent domains, not a full interface table. This leads to three direct results.
First, capabilities that are not registered are not visible to the client, and therefore cannot be called, reducing the space for 'try the model on the spot'. Second, the permission judgment has a unified mounting point - it is better to maintain by domain than to patch by a single interface. Third, when abnormal behavior occurs, it can be traced back to the specific intent domain, rather than just seeing a string of scattered calls.
For the content system, domains are usually grouped by read-write properties: reading class (document list, details, categories and tags) and writing class (save document, upload attachment, modify settings) are treated separately, and operation and maintenance class (backup, upgrade, multi-site) is grouped alone.
Within a round, the model often reads first and then makes several consecutive changes. There are two problems with item-by-item confirmation: first, people get tired and just click agree out of convenience after seeing the third one; second, when the batch changes are split for approval, the intermediate state is written into the database, making it difficult to determine which item caused the error after an error occurs.
The round-level approval increases the granularity by one level: the write operations within this round request are merged into one confirmation, and what the person sees when approving is an overall description of the batch changes. The write operations of this round are actually stored in the database after approval. Read operations do not participate in the approval to avoid wasting the confirmation cost on calls without side effects.
Station-level operations are not suitable for being casually called by external clients. AnQiCMS's MCP interface sets the exposure range according to the intent domain, the high-risk domain is closed by default, including backup, upgrade, multi-site management and other capabilities that affect the entire station, which must be explicitly enabled by the site itself to appear in the directory.
The recommended order at the beginning of the integration is: first, only open the read domain and run for a period of time to observe whether the calls are stable; then open the content writing domain and retain round-level approval; the operation domain should be opened individually as needed, and can be closed again after use. It is also recommended to integrate the batch capabilities of the Skills category at the same stage, because once batch operations are given to the client, the impact is much greater than that of saving a single article.
Should a new set of permissions be used when integrating external clients?No new account system is needed, but access credentials should be configured separately for the integration and opened minimally by domain; the rotation cycle of the credentials should be consistent with the back-end administrator password.
Does round-level approval slow down work?It will only add one confirmation. The real risk comes from batch overwrites when there is no confirmation, and the costs are not on the same level.
Does a read-only access require approval?No, read classes do not participate in approval; but if the read domain contains sensitive fields such as system settings, it should be tightened by domain rather than loosened for the entire group.
Preview
The cost of preventing collection interference codes is not the cost of plagiarism itself. It makes the copied text carry harmless noise to the original site but useless to the搬运方, thereby increasing the workload of cleaning and reformatting; it has limited effect on manual rewriting, abstract搬运, and only grabbing the title and abstract. The order of handling after being moved should be to first fix the first evidence, then check the对方抓取方式, and finally go through complaints and site ownership proof, rather than rely solely on technical interception.
Preview
Migration of more than a thousand historical articles can be done through two channels: content with a lot of text and images should be imported in batches using ZIP compression packages, and structured metadata and plain text content should be imported through Excel spreadsheets; then, to prevent duplicate entries, check for existing documents through the document import interface by title. Deleted official documents should go into the recycling bin for easy rollback. It is recommended to execute in batches, and check after each batch before scaling up.
Preview
The custom content model solves the problem of decoupling the content structure from business fields. When the article type is hardcoded, adding a parameter, scope, or attachment field requires changing the code and database structure; after changing to a configurable content model, fields can be defined in the background and can enter the editing page and list call, historical data is compatible with null values, and there is no need to go through a release cycle for a single structural change.
