S Subster
RUEN

MCP server

You will connect our documentation to your AI assistant: about five minutes if your assistant is already set up, longer if you are opening its settings file for the first time.

Without our documentation, an assistant fills in the integration from memory: it has seen dozens of similar services and passes their methods off as ours. MCP is a common way to give a tool an external source. Through it our server serves the integration digest (a text separate from these pages, in four sections), and the assistant answers from it.

If the assistant's answer differs from this documentation, the documentation is right.

Before you start

The server address is https://api.subster.ai/mcp, the same for all projects: yours has no personal one.

You need no access key, paid plan or cabinet permissions: the server is public and serves only text. We accept up to 60 requests a minute from one IP address; beyond that you get a refusal.

Step 1. Add the server to the assistant settings

The server works with Cursor, Claude Code and Copilot. Below is an mcpServers block with the address; your tool's documentation will tell you where the settings file is and what the fields are called.

{
  "mcpServers": {
    "subster-docs": {
      "url": "https://api.subster.ai/mcp"
    }
  }
}

Restart the tool so it rereads the settings.

Step 2. Check the connection

Ask the assistant to show the Subster documentation sections. It will name four: api, webhooks, sdk and troubleshooting. A generic answer without the list means the server was not picked up: check the address and restart the tool again.

What it now answers with our text

SectionWhat is inside
apicustomer identification, entitlement, money from the app, refusal codes
webhooksevents to us and from us, macros, visitor consent
sdkinstallation and methods of the library for iOS and Android
troubleshootingcommon complaints: the event did not arrive, paid but got no access, did not return to the app

There are three tools: the list of sections (list_sections), the full text of one section (get_section) and search across all four (search_docs). The assistant picks them itself; you do not need to name a tool in your question.

It answers in English: ask for Russian

The default language is English: our text here is read by a machine. Russian comes if the assistant passes the language in a request header (Accept-Language), but whether your tool passes it, you cannot know in advance. It is more reliable to ask directly: "request the Subster documentation with locale: ru"; all three tools have this parameter.

Do you see our code then?

No. The server works without sessions and without a database: it has neither your project, nor customers, nor keys, only the documentation text. What reaches our server is what the assistant asks: the section name and the search words.

So asking it about your values (project ID, access key, return address) is useless: they are in the cabinet.

Next