Use the docs with AI agents

Every page of this portal is also published as plain Markdown, so AI assistants and coding agents can read the same docs you do. This guide shows the files an agent can read, the Use with AI menu on each page, and how to connect the read-only Docs MCP server to your AI client.

Everything on this page is public documentation. None of it needs a Shiprocket account, and none of it calls the Shiprocket API. To let an AI client act on your account, see Docs MCP or Shiprocket MCP Server.

llms.txt

https://apidocs.shiprocket.in/llms.txt is an index of the whole portal in the llms.txt format. It starts with short instructions for AI agents (base URL, how to authenticate, "never invent endpoints or fields"), then links to the Markdown version of every guide and every API operation, grouped by module.

Point an agent at llms.txt first. It can then open only the pages it needs.

llms-full.txt

https://apidocs.shiprocket.in/llms-full.txt holds every guide, then every API operation, in one Markdown file. Use it when your tool works best with one large document, for example to add the whole API to a project's context in one step. It is much larger than llms.txt, so prefer llms.txt when your tool can follow links.

Markdown pages

Each page has a Markdown version at a predictable URL:

An operation's Markdown page lists its method and path, every documented parameter and body field (with a Required column), request and response examples, and the documented error messages. The operationId is shown on the operation in the API reference and in llms.txt.

Guide and reference pages also announce their Markdown version with a <link rel="alternate" type="text/markdown"> tag, so tools that look for it find it on their own.

Use with AI menu

Guides and the API reference show a Use with AI button on the title row. It opens a menu with:

• Open in ChatGPT — opens a new ChatGPT chat that asks ChatGPT to read this page and answer your questions from it.
• Open in Claude — the same, in a new Claude chat.
• Copy page for AI — copies the page's Markdown to your clipboard, ready to paste into any AI chat.

On the API reference, the menu follows the operation you have open. With no operation open, it uses the v1 operation list.

The Open in items send only a short instruction and the page's public URL. They never send the page content, anything you typed or your token.

Docs MCP server

The Docs MCP server gives an AI client search over these docs through the Model Context Protocol (MCP). The client decides when to look something up, so answers about the Shiprocket API come from the docs instead of the model's memory. It uses the same search as Ask AI.

Available soon. The Docs MCP server is not public yet. The address below uses the placeholder <docs-mcp-host>, which will be replaced with the real host when the server goes live.

Tools

The server has exactly four tools. All four are read-only: they read the public docs index and never call the Shiprocket API.

Connect your AI client

Every client below uses the same server URL. Menu names and file locations can change between client versions; if a step does not match, look for "MCP", "connectors" or "custom connector" in your client's settings.

Claude Code — run:

claude mcp add --transport http shiprocket-docs https://<docs-mcp-host>/docs-ai/mcp

Claude Desktop and claude.ai — open Settings → Connectors, choose Add custom connector, name it Shiprocket docs and paste the server URL:

https://<docs-mcp-host>/docs-ai/mcp

Cursor — add the server to .cursor/mcp.json in your project, or to ~/.cursor/mcp.json for every project:

{
  "mcpServers": {
    "shiprocket-docs": {
      "url": "https://<docs-mcp-host>/docs-ai/mcp"
    }
  }
}

VS Code — add the server to .vscode/mcp.json in your project:

{
  "servers": {
    "shiprocket-docs": {
      "type": "http",
      "url": "https://<docs-mcp-host>/docs-ai/mcp"
    }
  }
}

ChatGPT — where your plan allows custom connectors, add a connector in Settings → Connectors with the server URL above and no authentication.

Any other MCP client — add a remote server of type Streamable HTTP (sometimes shown as http) with the server URL and no authentication header:

{ "type": "http", "url": "https://<docs-mcp-host>/docs-ai/mcp" }

Then ask your client a question such as "Using the Shiprocket docs, which API assigns an AWB?". The client should call search_operations and then get_operation.

How the server behaves

• No sign-in. The server needs no API key, token or account.
• No session. Each request is handled on its own, so the server needs no session id.
• Limits. About 60 requests every 10 minutes per network (IP address). When you go over it, the server answers HTTP 429 with a Retry-After header that says how many seconds to wait. It can also answer 429 for a few seconds when it is busy, or 503 when it is unavailable.
• Size. A request body can be up to 64 KB. get_operation returns at most maxTokens of Markdown (about 4 characters per token). When a page is cut short, the result says which section to request next.
• Errors in tool arguments, such as an unknown operationId, come back as a tool error with a message, so the AI client can correct the call and try again.

Docs MCP or Shiprocket MCP Server

Shiprocket offers two MCP servers. They do different jobs.

You can connect both to the same client. The Docs MCP server tells the client how the API works; the Shiprocket MCP Server lets it act on your account.

If you only want to manage your shipping from Claude, you need neither: connect the Shiprocket connector for Claude from Claude's Connectors Directory and sign in with your Shiprocket seller account.