> For the complete documentation index, see [llms.txt](https://docs.autommate.com/userguide/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.autommate.com/userguide/mcp-and-ai-reference/read-documentation-from-an-ai-client.md).

# Read documentation from an AI client

Start with the AI entry page, choose the relevant task, then retrieve the exact tool reference and worked example. Use the smallest set of pages needed for the operation.

## Two connections with different purposes

| Connection                        | Purpose                                                                                                             |
| --------------------------------- | ------------------------------------------------------------------------------------------------------------------- |
| Autom Mate product MCP            | Exposes the authenticated product tools available to the connected account and environment.                         |
| Published documentation retrieval | Reads guides, contracts and examples. Reading documentation does not execute an Autom or grant product permissions. |

An external AI client's ability to open a web link depends on that client and its configured tools. A URL in a tool description is a useful pointer, not proof that the client fetched or applied the page.

## Read a specific page

GitBook serves a Markdown version of a published page by appending `.md` to its URL. For example:

```
https://docs.autommate.com/userguide/mcp-and-ai-reference/for-ai-agents-read-this-first.md
```

The site index is available at:

```
https://docs.autommate.com/userguide/llms.txt
```

Use the index to locate a page, then retrieve that page's Markdown. Reading the entire documentation corpus for a single tool call can introduce irrelevant or older guidance.

## Documentation MCP

GitBook documents a published-site MCP endpoint at:

```
https://docs.autommate.com/userguide/~gitbook/mcp
```

Use a compatible client's supported MCP connection procedure. This address is for documentation retrieval; it is separate from `https://mcp.autommate.app/mcp`, which exposes Autom Mate product operations.

First test initialization and one known-page retrieval. Client support and site access controls can affect the result. If access is refused, keep the failure visible and use permitted Markdown retrieval while the site administrator checks the access rule. Do not infer a successful connection from the endpoint's existence.

## Select the current reference

For the current MCP guide, start under:

```
https://docs.autommate.com/userguide/mcp-and-ai-reference/
```

Links containing `/userguide/userguide/` belong to the legacy documentation variant. The site-wide index can include both variants. Use the current MCP entry page and its explicit version scope for this catalog; consult legacy material only when investigating that documented legacy behavior.

If sources differ, compare the product build, connection, page scope and observation date. Read the actual tool schema before composing a call. Do not combine a legacy example with a newer tool's fields.

## What belongs in a tool description

Keep the following information close to the tool:

* Its purpose and important side effects.
* Required identifiers and how to discover them.
* Critical input rules and permission or tenant boundaries.
* Important retry or uncertainty behavior.
* A direct link to the exact tool reference.

Longer explanations, response examples, diagrams and related procedures can live in the documentation. For an execution tool, a link to the exact run guide is more useful than a generic documentation homepage.

## Check retrieval in your client

Before using documentation-assisted execution, perform a read-only check:

1. Ask the client to retrieve the AI entry page and name the page it actually read.
2. Ask it to find the `automs_run` reference and identify the supported trigger types.
3. Check that it explains the missing operation ID for webhook starts and the risk of a duplicate execution.
4. Ask what it would do if the required authoring tool were absent. It should identify that limit rather than invent a tool or schema.
5. Confirm that no product action was executed during this reading check.

If the client cannot retrieve the page, provide the exact relevant text or configure a supported documentation connection. Do not describe a generated answer as a successful retrieval test.

For GitBook's supported delivery formats and published documentation MCP, see [LLM-ready docs](https://gitbook.com/docs/getting-started/llm-ready-docs). A successful HTTP-level retrieval check does not establish that every AI client has been configured or tested.

## Suggested retrieval routes for service-authoring tools

These links are documentation routes for client and server implementers. Their publication does not automatically add links to product MCP tool descriptions.

| Before using                                                            | Retrieve                                                                                                                                         |
| ----------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------ |
| `services_create`, `services_release_version`, Work Item composition    | [Structured service authoring](https://docs.autommate.com/userguide/mcp-and-ai-reference/structured-service-authoring.md)                        |
| `deployments_create`, `deployments_update_draft`, `deployments_go_live` | [Deployment input resolution](https://docs.autommate.com/userguide/mcp-and-ai-reference/deployment-input-resolution.md)                          |
| Application/action/trigger discovery and Data Manager                   | [Native discovery](https://docs.autommate.com/userguide/mcp-and-ai-reference/native-app-trigger-and-data-discovery.md)                           |
| Switching scope or planning an unfamiliar operation                     | [Capabilities](https://docs.autommate.com/userguide/mcp-and-ai-reference/capabilities-and-version-scope.md)                                      |
| Error, missing schema or uncertain result                               | [Agent behavior](https://docs.autommate.com/userguide/mcp-and-ai-reference/agent-behavior-errors-missing-information-and-safe-interpretation.md) |

For each task, record the exact pages retrieved and the active tool declarations used. A useful read-only client check is to explain how a known Engage field and an earlier Work Item output satisfy two inputs without executing a service or asking the user to supply the same values again. If the output shape or field-path convention is unknown, the client should identify that gap explicitly.

Tool providers should include the matching canonical page URL alongside required fields, discovery prerequisites, context and side-effect notes. Keep critical constraints in the tool description itself so that a failed documentation fetch does not erase them. Updating these pages alone does not change the MCP server's descriptions or schemas.

## Browser WebMCP is a separate retrieval surface

On 21 October 2026, the published site exposed browser WebMCP page-read and documentation-search tools. A read of the deployment input-resolution page through that browser surface returned its full Markdown successfully. This confirms that specific browser retrieval; it does not establish that the remote documentation MCP endpoint initializes successfully in every client. Discover the tools offered by the current page rather than hardcoding generated tool names.
