# Set up your agent

**Give your agent the context it needs, then choose how it should work with Hlix.** It can read public documentation, use the CLI on your local project, or connect to your Workspace through MCP.

## Copy a setup prompt

Paste this into your assistant to help it discover the docs and choose the right tools for your request.

Use the current Hlix documentation to help me get set up.

Documentation:
- Start with the page index: https://docs.hlix.ai/llms.txt
- Read the setup guide: https://docs.hlix.ai/mcp/agent-setup.md
- For local CLI operations: https://docs.hlix.ai/reference/cli.md
- Fetch the full documentation only when needed: https://docs.hlix.ai/llms-full.txt
- For HTTP integrations, use the public contract: https://docs.hlix.ai/openapi.json
- Default API base for this environment: https://server.hlix.ai

Read the relevant pages as Markdown by using their .md URLs. If @hlix/cli is already installed, its bundled skills/hlix/SKILL.md also documents the commands and their inputs.

Workspace connection:
- Hosted MCP server: https://mcp.hlix.ai/mcp
- Documentation URLs explain the product. MCP accesses my Workspace; it does not provide documentation search.
- If MCP is connected, call hlix_whoami to confirm the Workspace and my access. Use hlix_projects and hlix_status before selecting an existing project.
- Discover the actual MCP tool schemas through tools/list and follow their documented inputs. If a tool is unavailable, explain what is needed.

This setup prompt does not authorize builds, purchases, or resource changes. Carry out only the work I request.

Connect MCP when your assistant needs Workspace access; public documentation can be read directly.

## Choose an entry point

| Your task | Use |
| --- | --- |
| Answer a product or setup question | The [documentation index](/llms.txt), followed by the relevant Markdown pages |
| Import or synchronize local files | The [CLI](/getting-started/install/) and its bundled skill |
| Inspect or manage cloud projects from an assistant | The [hosted MCP connection](/mcp/) |
| Write an API integration | The [OpenAPI document](/openapi.json), [API reference](/api/reference/) and [SDK guides](/sdk/) |

Use the Markdown URLs below for public documentation and MCP tools for Workspace actions.

## Read the docs as Markdown

Start with the page index and fetch only what the task needs:

```bash
curl -sS https://docs.hlix.ai/llms.txt
curl -sS https://docs.hlix.ai/reference/cli/auth.md
```

| Format | Purpose |
| --- | --- |
| [`/llms.txt`](/llms.txt) | Page index with descriptions to help an agent choose what to read |
| [`/llms-full.txt`](/llms-full.txt) | The full documentation for offline indexing or a large context window |
| A page's `.md` URL | One page without navigation chrome; for example, `/mcp/agent-setup.md` |
| [`/index.md`](/index.md) | The documentation homepage as Markdown |

Use **Copy Markdown** or the assistant actions on a page for the same content. The per-page path replaces the trailing slash with `.md`.

## Add the CLI skill

The public `@hlix/cli` npm package includes `skills/hlix/SKILL.md`, covering authentication, command discovery, JSON output and recovery. For an npm-global CLI installation, copy that guide into the skill directory your agent reads. This example uses `.agents/skills`:

```bash
mkdir -p .agents/skills/hlix
cp "$(npm root -g)/@hlix/cli/skills/hlix/SKILL.md" .agents/skills/hlix/SKILL.md
```

Use your assistant's documented skill directory if it differs. The [Claude Code plugin](/mcp/#claude-code) is another option: it includes the hosted MCP connection and a skill for the cloud project workflow.

Before using local project commands, have the agent run `hlix --help` and `hlix status` to inspect the installed commands, selected Workspace and folder binding. `hlix skill import` activates a resource in a cloud project; copying the CLI guide above equips the assistant running on your machine.

## Give it the API contract

The complete public contract is available without authentication:

```bash
curl -sS https://docs.hlix.ai/openapi.json
curl -sS https://server.hlix.ai/v1/api/openapi.json
```

Use the contract to inspect endpoint schemas and responses. For agency Workspace requests, the API key authenticates the user and `X-Organization-Id` selects the Workspace. Public and client-portal operations have their own requirements. See [API reference](/api/reference/) before generating calls.

## Work with the right project

After [connecting MCP](/mcp/), the agent should:

1. Use `hlix_whoami` to identify its Workspace and access.
2. Use `hlix_projects` to find the requested project, then `hlix_status` to inspect current work and questions.
3. Use `hlix_build` or `hlix_message` for the change you requested. If several projects match, use the returned candidates to resolve which one you meant.
4. Inspect `hlix_review` and `hlix_preview` when results are ready. Final merging remains your decision.

For connected provider tools, discover with `hlix_tools_search` and inspect the chosen schema with `hlix_tool_describe` before invoking it. Follow the [funding and retry rules](/mcp/tools/#durable-tool-calls) when recovering a provider call that stopped.

## Next steps

[Connect your AI](/mcp/)
  [MCP tools](/mcp/tools/)
  [CLI reference](/reference/cli/)