---
title: "Agents"
description: "What an agent needs to write and check an EZGHDeploy file and follow its deployments, the file's JSON Schema, llms.txt, and EZGH Cloud's MCP server."
---

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

# Agents

## The file's JSON Schema

`https://deploy.ezghcloud.com/schema/v1.json` is the JSON Schema of `EZGHDeploy`, and lists every
key the file accepts. Its `x-ezgh-attributes` lists the attributes each resource type offers to
references (`${{ resources.<name>.<attribute> }}`). An editor or agent can check a file against it,
but some rules span keys (a `web` service needs `port`, a reference must name a real resource), so
the authority is `ezgh deploy validate`, which checks the file locally, without a request, and
prints each problem with an example that would be accepted. The [file reference](/deploy/file-reference)
is generated from the same schema.

## These pages as text

`https://docs.ezghcloud.com/deploy/llms.txt` lists these pages, each also available as Markdown,
for an agent to read.

## EZGH Cloud's MCP server

The MCP server at `https://console.ezghcloud.com/mcp` reaches Deploy through its `call_api` tool,
with the service `deploy`, as you. `get_api_spec` with `deploy` describes its routes. For example:

| Task | Call |
| --- | --- |
| Check a file | `POST /v1/organizations/<org>/deploy/validate` with `{"file": "<the file>"}`, and `"environment": "<environment>"` to check one environment too |
| List stacks | `GET /v1/organizations/<org>/deploy/stacks` |
| List an environment's deployments | `GET /v1/organizations/<org>/deploy/stacks/<stack>/environments/<environment>/deployments` |
| Read a plan | `GET /v1/organizations/<org>/deploy/deployments/<deployment>/plan` |
| Approve that plan | `POST /v1/organizations/<org>/deploy/deployments/<deployment>/approve` with `{"planHash": "<the plan's hash>"}` |

An approval names the plan's hash, so it approves only the plan the agent read, and it needs the
same permissions as a person's: `deploy.deployments.approve` and every action in the plan.

Source: https://docs.ezghcloud.com/deploy/agents/index.mdx
