Help center MCP server
Let Claude, Cursor and other AI assistants search and read your help center over the Model Context Protocol. Business plan.
Every help center can run an MCP server. AI assistants that speak the Model Context Protocol can use it to search your published articles and read them as Markdown. With a token, they can also write drafts and update articles.
The MCP server is part of the Business plan.
Turn it on
Go to Help center → Settings → AI & search engines:
- Enable public MCP server gives read access to published articles. Anyone with the URL can use it, the same way anyone can read your public help center.
- Create access token adds write access. Treat the token like a password. It's shown only once, and Regenerate or Revoke token stop the old one immediately.
The server URL is:
https://asktrix.org/api/kb/YOUR_WORKSPACE_SLUG/mcp
The settings page shows the exact URL and a ready-made configuration.
Connect an assistant
Most MCP clients accept a configuration like this (Claude Desktop, Cursor, VS Code and others; the file name differs per client):
{
"mcpServers": {
"acme-kb": {
"type": "http",
"url": "https://asktrix.org/api/kb/acme/mcp"
}
}
}
For write access, add the token as a header:
{
"mcpServers": {
"acme-kb": {
"type": "http",
"url": "https://asktrix.org/api/kb/acme/mcp",
"headers": { "Authorization": "Bearer YOUR_ACCESS_TOKEN" }
}
}
}
Clients that can't send headers can pass the token as ?token=YOUR_ACCESS_TOKEN in the URL. Avoid that where you can, because URLs end up in logs.
Tools
| Tool | Access | Does |
|---|---|---|
search_help_center |
public | Finds articles for a question or keywords. Arguments: query (required), locale, limit (1–25, default 8). Returns id, title, slug, language, URL and an excerpt. |
get_article |
public | One article as Markdown, with URL, language, category and last update. Pass id, or slug with optional locale, or the article's url. |
browse_categories |
public | Every category of one language with the titles, ids and URLs of its articles (up to 50 each), plus all available languages. Argument: locale. |
create_article |
token | Creates an article from Markdown: title and content_markdown (required), description, locale, category_id, publish (default false, so it's a draft). |
update_article |
token | Changes an article by id. Only the fields you pass change: title, content_markdown, description, category_id, hidden, featured. publish: true publishes the new version, publish: false unpublishes. |
Without a token, the assistant only sees published, visible articles. With a token, drafts are visible too.
Protocol details
- Transport: Streamable HTTP.
POSTJSON-RPC 2.0 to the server URL and get a JSON response back. There's no server-sent event stream. - Methods:
initialize,ping,tools/list,tools/call.resources/listandprompts/listanswer with empty lists. - Batches of up to 20 requests are accepted.
- Rate limit: 120 requests per minute per IP address.
- CORS is open, so browser-based clients work.
- A
GETon the URL returns a short JSON description of the server.
Errors:
| JSON-RPC code | HTTP | Meaning |
|---|---|---|
-32001 |
401 | The access token is wrong |
-32001 |
200 | A write tool was called without a token (returned as a JSON-RPC error) |
-32003 |
403 | The public server is switched off, or the plan doesn't include it |
-32004 |
404 | No help center at this address |
-32005 |
429 | Rate limit exceeded |
-32700 |
400 | The body isn't valid JSON |
Try it with curl
curl -s https://asktrix.org/api/kb/acme/mcp \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_help_center","arguments":{"query":"reset password"}}}'