# Connect the Documentation MCP Server

Source: https://help.zira.us/developers/how-to-guides/connect-the-documentation-mcp-server
Summary: Add Zira's documentation to any MCP-compatible assistant so it can search the guides, API reference, and release notes while it works.
Updated: 2026-09-07

Zira publishes its documentation as an **MCP server**: a small service that an AI assistant can
call as a tool. Once it is connected, the assistant can search the product guides, the API
reference, and the release notes on its own, and answer from the current documentation instead of
from whatever it happens to remember about Zira.

Use it when you are building an integration with an AI coding assistant, or when you want an
assistant to answer product questions accurately.

## What you need

Nothing. The server is public, read-only, and needs no API key.

```text
https://help.zira.us/api/mcp
```

It serves the same documentation as this site, over the Model Context Protocol instead of over web
pages. It holds no customer data and cannot read or change anything in a Zira account.

## Connect it

The exact steps depend on your client, but every client needs the same two things: the transport
(`http`, sometimes called "streamable HTTP" or "remote") and the URL above.

### Claude Code

```bash
claude mcp add --transport http zira-docs https://help.zira.us/api/mcp
```

### Clients configured with a JSON file

Most desktop assistants and editors read a config file. Add one entry:

```json
{
  "mcpServers": {
    "zira-docs": {
      "type": "http",
      "url": "https://help.zira.us/api/mcp"
    }
  }
}
```

Claude Code picks this up in your next session; Claude Desktop reads its config at launch, so restart it.

### Anything else

Point any MCP client at the URL with the HTTP transport. If your client only supports local
`stdio` servers, put a stdio-to-HTTP bridge in front of it — the server itself needs no special
handling.

## Check that it worked

Ask the assistant something that can only be answered from the documentation, and watch for it to
call a tool:

```text
Using the Zira docs, how do I post readings by metric id?
```

A connected client lists three tools:

| Tool | What it does |
|---|---|
| `search_docs` | Searches the documentation and returns ranked matches with a snippet |
| `read_doc` | Returns one article in full, as Markdown |
| `list_docs` | Lists every article, so the assistant can see what exists |

Your client may also offer the articles as attachable **resources**, one per article, if it
supports them.

## What it can search

| Corpus | Contents |
|---|---|
| `guides` | Product guides — how operators, supervisors, and plant managers use Zira |
| `developers` | Developer documentation — authentication, endpoints, request and response shapes |
| `changelog` | Release notes, newest first |

The assistant picks the right corpus on its own, but you can steer it — "check the release notes"
or "look in the API reference" both work.

## Good to know

- **It is documentation, not data.** The server describes how Zira works. To read or write your
  own production data, use the [public API](/developers/api-reference/api-authentication), which
  does need an API key.
- **It is always current.** The server reads the same files this site is built from, so it reflects
  the documentation as published.
- **There is a fair-use limit.** Roughly a request a second per client. A normal assistant
  conversation never comes close; if you hit it, wait a minute.
- **Prefer it over pasting links.** An assistant with the server connected can find the right
  article itself, including ones you did not know existed.

## Related pages

- [Generate an API Key](/developers/how-to-guides/generate-an-api-key)
- [Authentication and Conventions](/developers/api-reference/authentication-and-conventions)