CatchCMS exposes admin operations to AI clients through MCP (Model Context Protocol). A business website team can divide maintenance into verifiable steps: connect a client, identify the site and content, save original values, submit specified fields, read the record back, and inspect the public pages. This tutorial uses the Chinese and English update to this tutorial itself to show actual argument structures.
1. Issue an access token and connect a client
A super administrator opens Content Management → MCP Access Tokens, selects an enabled admin account, and supplies a token name and validity period. The period is 1–365 days, with 90 days as the default. The full token appears once in the creation result. Save it immediately in the client's secure credential settings. The token list shows the owner, creation time, last use, and expiry.
Configure a client that supports Streamable HTTP and Bearer authentication:
- Transport:
Streamable HTTP. - Endpoint:
https://your-domain/api/mcp, using your own CatchCMS service domain. - Authorization:
Bearer <the full token copied at creation>.
After connecting, confirm the server information and availability of search and execute. The token inherits its owner's functional permissions and site assignments. Expiry, revocation, account suspension, or account deletion ends access. Revoke tokens as needed when tasks end or staff access changes.
2. Discover each operation and its current schema
Use search before each operation. Read the returned operation_id, site_required, and input_schema. These are arguments for a search that finds the content-list operation:
{
"query": "cmsContentsIndex",
"limit": 1
}
Then use execute with the fixed operation ID discovered by search. Place operation_id, site_id, language, path, query, and body at the top level as required by that operation's schema. These fields directly form the execute tool argument object. Adding an extra parameters, arguments, or input layer inside this object causes a structure error. A raw MCP JSON-RPC tools/call request uses the protocol-defined params.arguments to hold the entire tool argument object. The catalog's input_schema describes path, query, and body. Operations with site_required=true require top-level site_id. When it is false, consult the specific operation rules for optional context. Platform operations such as the site list use their business arguments and omit top-level site_id and language. Top-level language sets the request language context; persistent English updates use body.translations.en. This example uses en, the actual code/locale of the English language bound to this website. For another site, use its enabled, bound language key. The current schema governs execution.
3. Read the real site, model, and target record
Search for cmsSitesIndex, then execute it to list sites available to the account:
{
"operation_id": "cmsSitesIndex"
}
The official site record read on October 6, 2026 has site_id=1, domain cms.catchadmin.vip, and enabled Chinese and English languages. Next, discover and execute cmsContentsSchema with top-level site_id: 1 and query: {"model_id": 1}. Confirm model 1, “Official Website Content,” and category 1, “Product Features.” These IDs come from this website's current records; use the IDs returned for your own site.
After discovering cmsContentsIndex, locate this tutorial by model, category, and its existing slug:
{
"operation_id": "cmsContentsIndex",
"site_id": 1,
"query": {
"model_id": 1,
"category_id": 1,
"slug": "mcp-admin-operation-guide"
}
}
The target returned by the list has content ID 53 and the Chinese title “CatchCMS MCP 使用教程:连接 AI 并完成内容更新.” The slug filter is a partial-match search, so check each result. The list requires query.model_id. Use the real content ID returned by the list as path.content. After searching for cmsContentsShow, read the detail:
{
"operation_id": "cmsContentsShow",
"site_id": 1,
"path": {
"content": 53
}
}
Save the request, original response, Chinese body and description, and English translations as a task backup. Also retain the titles, slug, publishing state, published_at, model, category, and field_values.schema_type for comparison. Verify the target using the request's site_id and path.content, together with the returned slug and title. The data in this detail response omits the content ID.
4. Review copy and update the specified bilingual fields
Check product claims, operation arguments, and links, then verify that the English content fully corresponds to the Chinese version. Define the allowed fields, search for cmsContentsUpdate, and read its partial-update and translation-merge rules. This example submits only the Chinese and English SEO descriptions. The text matches the descriptions used in this tutorial's update:
{
"operation_id": "cmsContentsUpdate",
"site_id": 1,
"path": {
"content": 53
},
"body": {
"seo_description": "CatchCMS MCP 实操教程:签发访问令牌,以 Streamable HTTP 与 Bearer 接入客户端,通过 search 获取 schema,在真实官网内容中更新中英文正文和 SEO 描述,保存原值、回读比对并验收公开页面。",
"translations": {
"en": {
"seo_description": "Connect an MCP client to CatchCMS with Streamable HTTP and Bearer authentication. Follow a real bilingual tutorial update: discover schemas, back up content, update copy and SEO descriptions, read back, and check public pages."
}
}
}
}
The description example is a subset of the full update arguments. The complete update also submits the Chinese body in body.content and the English body in body.translations.en.content. The scope is body copy and descriptions in both languages. The existing bilingual titles, slug, keywords, publication timestamp, publishing state, and schema_type=TechArticle are preserved. Partial updates merge by field: omitted fields keep their current values, while explicit null and empty strings save the supplied values.
For your own record, a description update provides a small first exercise. Prepare exact copy and the target record each time, and execute within the user's authorization. The business owner supplies and reviews product materials, deployment requirements, and customer case studies.
5. Read back and inspect the public pages
Check the MCP response's isError and structuredContent: a business error can appear in an HTTP 200 response. A successful content update returns data: true. Then execute the same cmsContentsShow request from step 3 again. Compare the Chinese body, seo_description, translations.en.content, and English description against the submitted values, then check the fields being preserved. Keep the exact update arguments, execution response, and readback alongside the original values.
This tutorial's public URLs are the Chinese tutorial and the English tutorial. After updating, check them anonymously using this checklist:
- Both pages return HTTP 200. The original titles are preserved, each meta description matches its updated language value, and the body shows the submitted changes.
- Each canonical points to the actual URL for that language. Language links open the corresponding Chinese and English pages, and the declared page language matches the content.
- Relevant body links are accessible and match the language and context. This tutorial retains TechArticle structured data; inspect other applicable runtime output, including BreadcrumbList.
- At a mobile width of approximately 390 pixels, review paragraphs, code blocks, links, and page width. Confirm readable JSON and controlled horizontal overflow. The code blocks allow wrapping and independent horizontal scrolling.
Retain separate evidence for saved content, matching readback, and accessible public pages. Observe search indexing, rankings, actual citations in AI answers, and inquiries after publication. Those outcomes require real measurements and citation evidence.
6. Permissions and errors
search returns operations permitted for the account. execute checks functional permissions and site scope on every call, and writes produce MCP operation logs. Retain permission or validation errors, check the token owner, site assignments, and schema, then correct the request. Theme uploads are trusted code deployment capabilities and should be assigned to staff responsible for development and deployment.
7. Apply the workflow to regular maintenance
This page covers connection, arguments, and verification steps. MCP content management explains product capabilities and suitable tasks. Continue with MCP data scope, permissions and security, AI theme development, and AI theme tools. Complete one record's read, backup, update, readback, and public verification before applying the same workflow to ongoing business website maintenance.
Plan the next step for your project
Review source-code plans and updates, then use project enquiries to confirm site counts, languages, deployment and client-delivery needs.