---
title: "Use these docs from an agent"
description: "Point a coding agent at llms.txt, a page's Markdown copy, or the docs' MCP server."
---

> Documentation Index
> Fetch the complete documentation index at: https://docs.oxabl.org/llms.txt
> Use this file to discover all available pages before exploring further.

# Use these docs from an agent

These docs are published for agents as well as people, in three forms. Use whichever your agent supports.

| Form | Where | Use it when |
| --- | --- | --- |
| An index | `/llms.txt` | The agent can fetch URLs. It lists every page with a one-line description and a link to its Markdown. |
| One page as Markdown | the page's URL plus `index.md`, such as `/toolchain/commands/check/index.md` | The agent needs one page, without the site around it. Every page has a **View as Markdown** link. |
| Everything at once | `/llms-full.txt` | The agent should read the whole site in one request. |
| An MCP server | `/mcp` | The agent speaks the Model Context Protocol. It needs no key. |

## Connect to the MCP server

The server speaks Streamable HTTP at `/mcp`. Add it to your agent the way its docs describe for a remote MCP server; the **Copy MCP URL** link on every page copies the address. It's read-only and has two tools:

- **`list_pages`** lists pages with their title, URL, Markdown URL, description and section. Pass `section` to list one product: `toolchain` or `cache`.
- **`get_page`** returns one page as Markdown. Pass the `url` that `list_pages` gave, or a path such as `/toolchain/commands/check/`.

There's no search tool. The agent lists a section, picks a page by its description, and reads it.

To try the tools by hand, use the MCP Inspector. This run is against a local copy of the site (`npm run dev` in the docs repository); against the published site, use its address. The URLs in the answer are always the published ones.

```sh
npx @modelcontextprotocol/inspector --cli http://localhost:4321/mcp \
  --transport http --method tools/call --tool-name list_pages --tool-arg section=cache
```

```json
{
  "content": [
    {
      "type": "text",
      "text": "{\"pages\":[{\"title\":\"What the cache is for\",\"url\":\"https://docs.oxabl.org/cache/\",\"markdownUrl\":\"https://docs.oxabl.org/cache/index.md\",\"description\":\"What an oxabl cache does for a team and for CI, what it can and can't see, and what runs today.\",\"section\":\"cache\"},{\"title\":\"oxabl-cache-admin\",\"url\":\"https://docs.oxabl.org/cache/operate/admin/\",\"markdownUrl\":\"https://docs.oxabl.org/cache/operate/admin/index.md\",\"description\":\"Every operator command of oxabl-cache-admin, with its flags, output and exit codes.\",\"section\":\"cache\"},{\"title\":\"Self-host\",\"url\":\"https://docs.oxabl.org/cache/self-host/\",\"markdownUrl\":\"https://docs.oxabl.org/cache/self-host/index.md\",\"description\":\"Run an oxabl cache on your own machines.\",\"section\":\"cache\"},{\"title\":\"Self-host quick start\",\"url\":\"https://docs.oxabl.org/cache/self-host/quick-start/\",\"markdownUrl\":\"https://docs.oxabl.org/cache/self-host/quick-start/index.md\",\"description\":\"Run an oxabl cache on one machine with the container image, create the first tenant, and check that it's healthy.\",\"section\":\"cache\"}]}"
    }
  ]
}
```

## How an agent uses it

Say a CI job fails and its log ends with `oxabl check` exiting `2`. An agent with the MCP server connected can:

1. Call `list_pages` with `section: "toolchain"` and find **oxabl check** by its description.
2. Call `get_page` on it and read the exit code table: `2` is a usage or configuration error, such as an `oxabl.toml` it can't use.
3. Call `get_page` on **oxabl.toml reference** for the key the error names, fix the file, and run the check again.

The agent reads the same pages you do, from the same build, so its answer matches the docs for the version that's published.

Source: https://docs.oxabl.org/agents/index.mdx