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

# MCP, llms.txt, and skill.md

> Connect AI tools to the MTA API docs through the hosted MCP server, llms.txt / llms-full.txt files, and an installable Claude skill — all generated automatically.

These docs are built to be consumed by AI tools, not just people. Every page is available in machine-readable form, plus there's a hosted MCP server and an installable Claude skill — all generated and kept up to date automatically, at no extra cost. Point your agent at any of the endpoints below to give it grounded, current knowledge of the `mta-js` SDK and the MTA API.

<CardGroup cols={2}>
  <Card title="MCP server" icon="plug" href="#mcp-server">
    Connect Cursor, Claude, Codex, VS Code, and other MCP clients to search these docs live.
  </Card>

  <Card title="llms.txt" icon="file-text" href="#llms-txt">
    A machine-readable index of every page, for AI tools to discover and navigate.
  </Card>

  <Card title="skill.md" icon="sparkles" href="#claude-skill">
    Install these docs as a Claude skill in Claude Code and Claude Desktop.
  </Card>

  <Card title="Markdown pages" icon="markdown" href="#markdown-for-any-page">
    Append `.md` to any docs URL to get the raw Markdown of that page.
  </Card>
</CardGroup>

## MCP server

The docs are exposed as a hosted [Model Context Protocol](https://modelcontextprotocol.io) (MCP) server, so AI clients can search and read the documentation as a tool — answering questions about the SDK and API with up-to-date context. It exposes a **search** tool and a **filesystem-style navigation** tool over the docs.

**Server URL**

```text theme={null}
https://mtaapi.dev/docs/mcp
```

<Tabs>
  <Tab title="Cursor">
    Add the server to `~/.cursor/mcp.json` (global) or `.cursor/mcp.json` in your project:

    ```json theme={null}
    {
      "mcpServers": {
        "mta-api": {
          "url": "https://mtaapi.dev/docs/mcp"
        }
      }
    }
    ```
  </Tab>

  <Tab title="Claude Code">
    Add the remote MCP server from your terminal:

    ```bash theme={null}
    claude mcp add --transport http mta-api https://mtaapi.dev/docs/mcp
    ```
  </Tab>

  <Tab title="Codex">
    Add the server to `~/.codex/config.toml`:

    ```toml theme={null}
    [mcp_servers.mta-api]
    url = "https://mtaapi.dev/docs/mcp"
    ```

    Or add it from your terminal:

    ```bash theme={null}
    codex mcp add mta-api --url https://mtaapi.dev/docs/mcp
    ```

    Remote (streamable HTTP) MCP servers require Codex's RMCP client. If the server isn't picked up, enable it in `~/.codex/config.toml`:

    ```toml theme={null}
    experimental_use_rmcp_client = true
    ```
  </Tab>

  <Tab title="Claude / Claude Desktop">
    In **Settings → Connectors**, choose **Add custom connector**, then paste the server URL:

    ```text theme={null}
    https://mtaapi.dev/docs/mcp
    ```
  </Tab>

  <Tab title="VS Code">
    Add the server to `.vscode/mcp.json`:

    ```json theme={null}
    {
      "servers": {
        "mta-api": {
          "type": "http",
          "url": "https://mtaapi.dev/docs/mcp"
        }
      }
    }
    ```
  </Tab>
</Tabs>

Once connected, ask your assistant things like *“How do I fetch a subway arrival board with mta-js?”* and it will pull answers straight from these docs. A discovery endpoint is also published at `/.well-known/mcp` so compatible tools can locate the server automatically.

## llms.txt

Following the [llms.txt standard](https://llmstxt.org), two machine-readable files are generated automatically and stay in sync with these docs:

| File            | URL                                                                              | Contents                                                                                    |
| --------------- | -------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------- |
| `llms.txt`      | [`https://mtaapi.dev/docs/llms.txt`](https://mtaapi.dev/docs/llms.txt)           | A structured index of every page with titles, links, and descriptions — a sitemap for LLMs. |
| `llms-full.txt` | [`https://mtaapi.dev/docs/llms-full.txt`](https://mtaapi.dev/docs/llms-full.txt) | The entire documentation site concatenated into one file for direct context loading.        |

Use `llms.txt` when a tool should discover and fetch only the relevant pages, and `llms-full.txt` when you want to drop the whole corpus into a model's context window.

## Claude skill

These docs are also published as an installable [Claude skill](https://www.anthropic.com/news/skills), described by a `skill.md` manifest that summarizes the SDK's capabilities, methods, and best practices for agents:

```text theme={null}
https://mtaapi.dev/docs/skill.md
```

Add it to any agent that supports the [agent skills](https://agentskills.io) spec with the skills CLI:

```bash theme={null}
npx skills add https://mtaapi.dev/docs
```

This loads the MTA API capabilities into the agent's context so it can reference the guides and reference material on demand while you work. MCP-compatible tools can also read the `skill.md` content directly.

## Markdown for any page

Every page is available as raw Markdown — just append `.md` to its URL. For example:

```text theme={null}
https://www.mtaapi.dev/docs/guides/subway-arrivals.md
```

This is handy for piping a single page into a prompt or script without any HTML.

<Note>
  All of the above — the MCP server, `llms.txt`, `llms-full.txt`, `skill.md`, and per-page Markdown — are generated and hosted automatically. There's nothing to configure and no extra cost.
</Note>