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

> ## Agent Instructions
> To act on a user's behalf in Guild, use the Guild REST API at https://api.guild.ai/v1.
> The OpenAPI spec is unauthenticated: https://docs.guild.ai/openapi.json (or openapi.yaml).
> Authenticate with an account API key, sent as HTTP Basic (key id as username, secret as password) or as `Authorization: Bearer <api_key_id>:<api_key_secret>`. See https://docs.guild.ai/api-reference/introduction#authentication.
> To run one agent on demand without account-wide access, use an API trigger instead: https://docs.guild.ai/platform/api-triggers.
> To act on Guild from an MCP client, connect to https://api.guild.ai/mcp (OAuth) or run `guild mcp` locally. Setup: https://docs.guild.ai/api-reference/mcp.
> To search these docs from an MCP client, connect to https://docs.guild.ai/mcp. The full page index is https://docs.guild.ai/llms.txt.

# Run Steel browser sessions from Guild agents

> Connect Guild agents to Steel to scrape pages, capture screenshots, render PDFs, manage cloud browser sessions, and send computer-use actions.

Integrations · Steel

Guild agents use Steel to open isolated cloud browsers, scrape pages, capture screenshots, render PDFs, and send computer-use actions against live sessions. Guild stores the Steel API key server-side so agent code never sees it. Credential policies let you restrict which operations each agent can call, and every request runs through Guild's audit trail.

> [card] [Book a demo](https://guild-ai.cal.com/gtm/meeting-with-guild.ai)
>
> See how teams govern agents that use Steel.

> [card] [Talk to sales](https://guild.ai/talk-to-sales)
>
> Get access to Guild, then connect Steel to your first agent.

## What the Steel integration exposes

The integration provides tools to create and release browser sessions, list and inspect them, and retrieve live viewer URLs and open pages. Separate tools scrape a URL and return content with metadata and links, capture a hosted PNG, render a hosted PDF, retrieve cookies and storage for a session, and send a single computer-use action against a running session.

## How it works

**1. Open credentials in Guild**

Go to Access & setup > Credentials in Guild. Click Connect, select steel, and paste your Steel API key into the API Key field.

**2. Import and wire the tools**

Import SteelTools from the guildai\~steel package and spread it into your agent's tools object. Guild sends the key as the steel-api-key header on every request.

**3. Select only what you need**

Tools use the steel\_ prefix. Use pick() from the Agent SDK to include only the specific tools your agent requires, as documented on the [tool sets page](https://docs.guild.ai/sdk/tools.md#selecting-specific-tools).

## Why Steel session access needs governing

Steel tools can launch browsers, scrape arbitrary URLs, capture screenshots, render PDFs, retrieve session cookies and storage, and send computer-use actions against live sessions. Guild stores the API key server-side through its credential proxy, so agent code never handles it directly. You can scope a credential to one workspace, one member, or one agent so access stays as narrow as the job requires. See [Credentials](https://docs.guild.ai/platform/credentials.md) for scoping options.

Credential policies let you write ALLOW and DENY rules that the proxy enforces before any request reaches Steel. You can restrict an agent to read-only operations like steel\_scrape and steel\_screenshot while blocking session creation or computer actions entirely. A DENY rule always overrides a matching ALLOW rule. Agents cannot evaluate or bypass policies themselves. Without scoped rules the credential falls back to allow-all; delete that default once your rules cover everything agents should reach. See [Credential policies](https://docs.guild.ai/platform/credential-policies.md) for rule syntax.

## About the Steel integration

This integration lets agents launch and manage isolated Steel cloud browser sessions. Agents can scrape web content, capture screenshots, render PDFs, and inspect session state including cookies and storage. They can also send computer-use actions against running sessions. The integration is `guildai~steel`.

## Authentication

* **Type:** API key
* **Header:** Guild sends the key as `steel-api-key: <key>`

### Setup

**1. Connect Steel**

In Guild, go to **Access & setup > Credentials**, click **Connect**, select **steel**, and paste the key into **API Key**.

See [Credentials](https://docs.guild.ai/platform/credentials.md) for workspace, member, and agent-scoped credentials.

## Usage

```typescript theme={null}
import { SteelTools } from "@guildai-services/guildai~steel"
import { llmAgent } from "@guildai/agents-sdk"

export default llmAgent({
  systemPrompt: "You work with Steel.",
  tools: { ...SteelTools },
})
```

### Selecting specific tools

Tools are named with the `steel_` prefix. Use [`pick()`](https://docs.guild.ai/sdk/tools#selecting-specific-tools) to include only what your agent needs.

## Tools

Tools in the latest published version of `guildai~steel`:

| Tool | What it does |
| - | - |
| `steel_computer` | Send one computer-use action against an existing session. |
| `steel_create_session` | Launch an isolated Steel browser. |
| `steel_get_session` | Get one session by id |
| `steel_get_session_context` | Get cookies and storage for a session |
| `steel_get_session_live_details` | Get live viewer URLs and open pages |
| `steel_list_sessions` | List browser sessions |
| `steel_pdf` | Render a hosted PDF of a URL |
| `steel_release_all_sessions` | Release all live sessions |
| `steel_release_session` | Release one browser session |
| `steel_scrape` | Load a URL in a fresh Steel browser and return content, metadata, and links. |
| `steel_screenshot` | Capture a hosted PNG of a URL |

## Frequently asked questions

> [accordion] Can an agent retrieve cookies and local storage from a running Steel session?
>
> Yes. The steel\_get\_session\_context tool returns cookies and storage for a session identified by its id.

> [accordion] Does Guild expose webhook events from Steel?
>
> No. The catalog entry for guildai\~steel lists no webhook events, so event-triggered workflows based on Steel activity are not available.

> [accordion] Can I limit one agent to scraping only, without letting it launch or release sessions?
>
> Yes. Write a credential policy that ALLOW-lists steel\_scrape and steel\_screenshot and DENY-lists steel\_create\_session, steel\_release\_session, and steel\_release\_all\_sessions, then scope the policy to that agent. See [Credential policies](https://docs.guild.ai/platform/credential-policies.md).

> [accordion] How does Guild authenticate requests to Steel?
>
> Guild sends the key as the steel-api-key header on every request. The key is stored as a Guild credential and injected server-side by the credential proxy; the agent never sees it.

## Put a governed agent on Steel this week

> [card] [Book a demo](https://guild-ai.cal.com/gtm/meeting-with-guild.ai)
>
> See how teams govern agents that use Steel.

> [card] [Talk to sales](https://guild.ai/talk-to-sales)
>
> Get access to Guild, then connect Steel to your first agent.

## Related

**Also runs on Guild:** [Browserbase](https://docs.guild.ai/integrations/browserbase.md) · [Cloudflare](https://docs.guild.ai/integrations/cloudflare-openapi.md) · [Firecrawl](https://docs.guild.ai/integrations/firecrawl.md) · [Kernel](https://docs.guild.ai/integrations/kernel.md)

**Read:** Docs: [Credential policies](https://docs.guild.ai/platform/credential-policies.md) · [Credentials](https://docs.guild.ai/platform/credentials.md) · [Tools in agents](https://docs.guild.ai/sdk/tools.md)
