Context7

How AI coding assistants can pull current marzban-sdk and marzban-mcp documentation on demand, instead of answering from stale training data.

This is different from the MCP server (marzban-mcp). That one gives an AI agent operational access to a running Marzban panel — creating users, restarting the core. Context7 gives any AI coding assistant read access to this documentation, kept current with every release.

What it is

Context7 is an MCP server that resolves a library name to its current documentation and code examples, then feeds them into an AI assistant's context — instead of the assistant answering from training data that may be a year old. It exposes two tools: resolve-library-id (name → library ID) and get-library-docs (library ID → current docs). Most clients trigger it with the phrase use context7 in a prompt.

marzban-sdk is already indexed:

Library ID: /websites/ilmar7786_github_io_marzban-sdk
Source: this documentation site

Why it matters here

This monorepo ships releases often, and an assistant's training data doesn't move at the same pace — a model can confidently suggest an API shape that changed two versions ago. Context7 closes that gap automatically: your assistant re-fetches current docs on every relevant question, without anyone pasting a page into the chat by hand.

For a team evaluating this SDK, it's a concrete signal: the project is set up to be used correctly by AI-assisted developers, not just documented for humans reading top to bottom.

Two ways an AI can read these docs

Mechanism	How a model gets it	Best for
`llms.txt` / `llms-full.txt`	A direct URL any assistant can fetch — no MCP required	Clients without MCP support, or pasting a link straight into a chat
Context7 MCP	Automatic, inside your IDE, for every library you use — not just this one	Day-to-day development in an MCP-aware editor

How a request flows

You ask your assistant a question and mention use context7 (or your client's equivalent).
The assistant calls Context7's resolve-library-id for "marzban-sdk" — or you supply the ID above directly.
Context7 calls get-library-docs, pulling from this site's current content.
The docs come back into the assistant's context.
The answer reflects the version you actually have installed, not whatever the model last saw in training.

Setting it up

Context7 runs either as a remote HTTP server (https://mcp.context7.com/mcp, no install) or locally via npx -y @upstash/context7-mcp. An API key from context7.com is optional but raises your rate limit.

Add to .cursor/mcp.json (project) or ~/.cursor/mcp.json (every project):

{
  "mcpServers": {
    "context7": {
      "url": "https://mcp.context7.com/mcp"
    }
  }
}

For any other MCP client, see Context7's own client list — the shape is almost always one of the two above.

Then just ask, mentioning the trigger phrase:

Add pagination to sdk.user.getUsers using paginateAll — use context7 for marzban-sdk's current API

Or skip the resolve step and reference the library directly:

use library /websites/ilmar7786_github_io_marzban-sdk for marzban-sdk and marzban-mcp

In practice

Feature work — ask for a change against a specific module ("add X to sdk.node") and get an answer against the API you actually have installed, not a remembered one.
Onboarding — a new hire's assistant is correct about this SDK from day one, without anyone walking them through what changed since the model's training cutoff.
Agentic workflows — an autonomous coding agent resolves documentation mid-task on its own, the same way it would for any other dependency it needs current facts about instead of guessing.