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

# Use these docs with AI tools

> Machine-readable exports, an MCP server, and one-click handoffs so your AI assistant can read the Upwell docs while you build.

# Use these docs with AI tools

Most integrations against Upwell are now written with an AI assistant in the loop. This
site ships machine-readable surfaces so that assistant reads the **same verified truth**
you do — including the per-endpoint retry and delete semantics baked into the
[API reference](/api-reference/introduction) and the generated
[vocabulary](/api-reference/vocabulary).

<Note>
  These surfaces cover the **public** docs. Pages that require authentication are not
  included in `llms.txt`, `llms-full.txt`, or the Markdown exports — those are built from
  the public pages only.
</Note>

## Point your assistant at the docs

<CardGroup cols={2}>
  <Card title="llms.txt" icon="list" href="https://docs.upwell.com/llms.txt">
    A machine-readable index of every public page — the sitemap your AI tool reads first.
    Also served at `/.well-known/llms.txt`.
  </Card>

  <Card title="llms-full.txt" icon="file-lines" href="https://docs.upwell.com/llms-full.txt">
    The full content of the public docs in one file, for tools that ingest everything up
    front instead of following links.
  </Card>
</CardGroup>

Every page is also available as plain Markdown — append `.md` to its URL
(for example, `docs.upwell.com/authentication.md`), or press
<kbd>Cmd</kbd>+<kbd>C</kbd> (<kbd>Ctrl</kbd>+<kbd>C</kbd> on Windows) on any page to copy
it as Markdown for pasting into a conversation.

## The contextual menu

The menu at the top of every page bundles these handoffs into one click: copy the page as
Markdown, view it raw, open it as context in **Claude** or **ChatGPT**, connect the docs
MCP server to **Cursor** or **VS Code**, or copy the MCP server URL. On API reference
pages it can also download the OpenAPI spec directly.

## Connect the MCP server

The docs host a search [MCP](https://modelcontextprotocol.io) server at:

```text theme={null}
https://docs.upwell.com/mcp
```

Once connected, your AI tool can search and retrieve doc content live while you work —
ask it *"is `POST /api/rest/carriers` retry-safe?"* and it answers from the same
[idempotency contract](/conventions/idempotency-and-retries) documented here, instead of
guessing.

<Tabs>
  <Tab title="Claude Code">
    ```bash theme={null}
    claude mcp add --transport http upwell-docs https://docs.upwell.com/mcp
    ```
  </Tab>

  <Tab title="Claude">
    1. Open [Settings → Connectors](https://claude.ai/settings/connectors).
    2. Click **Add custom connector**.
    3. Name it `Upwell docs` and paste `https://docs.upwell.com/mcp`.
    4. In a conversation, click the **+** (attachments) button and select the connector.
  </Tab>

  <Tab title="Cursor / VS Code">
    Use **Connect to Cursor** or **Connect to VS Code** in the contextual menu at the top
    of any page — both install the server in one click.
  </Tab>
</Tabs>

Agents and tools can also discover the server without configuration via the standard
discovery document at `/.well-known/mcp`.

## skill.md

For agents that consume [skill files](https://agentskills.io), an automatically generated
capability summary of the Upwell platform is available at
[`docs.upwell.com/skill.md`](https://docs.upwell.com/skill.md). It stays in sync with the
docs and requires no maintenance; agents connected to the MCP server can also discover it
as an MCP resource.

## What to rely on

AI answers are only as good as their source. Two properties of these docs are worth
knowing when you delegate to an assistant:

* **Generated, not typed** — enum values on the [vocabulary page](/api-reference/vocabulary)
  and the endpoint schemas in the reference come from Upwell's live OpenAPI pipeline, so an
  assistant reading them can't inherit hand-typo'd values.
* **Behavioral contracts live on the endpoint pages** — where an endpoint has a non-obvious
  contract (retry safety, delete semantics, in-`200` business errors), it's stamped onto
  that generated endpoint page, so landing there from search surfaces it.

If an assistant's claim about the API contradicts these docs, trust the docs — and if the
docs turn out wrong, [tell us](mailto:support@upwell.com); that's a bug we fix at the
source.