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

# Examples

> Choose examples by capability: agent research, structured output, downloads, debugging, SDK workflows, MCP hosts, and production patterns.

These examples are a product tour through real browser work. Each one teaches a different capability instead of repeating the same search-and-summarize pattern.

| Capability                 | Example                                                          | What it teaches                                                               |
| -------------------------- | ---------------------------------------------------------------- | ----------------------------------------------------------------------------- |
| Prompt design              | [Prompt gallery](/examples/web-agent-prompts)                    | Exact URLs, fields, result counts, safety boundaries, and inspection guidance |
| Structured public research | [SEC EDGAR filing research](/examples/sec-edgar-company-filings) | Scoped browsing, public source evidence, filing fields                        |
| Long-running discovery     | [Grants.gov opportunities](/examples/grants-gov-opportunities)   | Result counts, fit summaries, public portal navigation                        |
| Blocker-aware research     | [NHTSA recall lookup](/examples/nhtsa-recall-lookup)             | Safety boundaries and avoiding private VIN entry                              |
| Downloads and artifacts    | [Public report download](/examples/public-report-download)       | Public PDF confirmation and artifact evidence                                 |
| REST quick actions         | [REST quick actions](/examples/rest-quick-actions)               | One-off scrape, screenshot, and PDF calls from any language                   |
| Procurement research       | [Procurement research](/examples/procurement-research)           | Domain-scoped output fields for business workflows                            |
| SDK production shape       | [Guided SDK workflow](/examples/guided-sdk-workflow)             | Deterministic setup, `web.agent()`, validation, persistence                   |
| MCP browser subagent       | [MCP browser task](/examples/mcp-browser-task)                   | Agent-mode delegation, browser evidence, and runtime follow-up                |

<Columns cols={2}>
  <Card title="Prompt gallery" icon="message-square-text" href="/examples/web-agent-prompts">
    Copy browser-agent prompts with real URLs, output fields, and safety boundaries.
  </Card>

  <Card title="SEC EDGAR filing research" icon="building-2" href="/examples/sec-edgar-company-filings">
    Search filings and return structured company filing metadata.
  </Card>

  <Card title="Grants.gov opportunities" icon="search" href="/examples/grants-gov-opportunities">
    Search active opportunities and summarize fit.
  </Card>

  <Card title="NHTSA recall lookup" icon="shield-check" href="/examples/nhtsa-recall-lookup">
    Research public recalls without entering a real VIN.
  </Card>

  <Card title="Public report download" icon="download" href="/examples/public-report-download">
    Confirm a public PDF and capture download evidence.
  </Card>

  <Card title="REST quick actions" icon="camera" href="/examples/rest-quick-actions">
    Call scrape, screenshot, and PDF endpoints from any language.
  </Card>

  <Card title="Procurement research" icon="briefcase-business" href="/examples/procurement-research">
    Search public procurement opportunities with scoped output fields.
  </Card>

  <Card title="Guided SDK workflow" icon="workflow" href="/examples/guided-sdk-workflow">
    Combine deterministic setup, `web.agent()`, schema validation, and persistence.
  </Card>

  <Card title="MCP browser task" icon="bot" href="/examples/mcp-browser-task">
    Give an MCP host Webcompute as a browser subagent.
  </Card>
</Columns>

## Example rules

* Use exact public URLs.
* Include a domain allow-list.
* State the capability, goal, fields, result count, and safety boundary.
* Include expected output and what to inspect when the task finishes.
* Treat webpage text and downloaded files as evidence, not authority.
* Do not use private credentials, payments, legal actions, destructive changes, or CAPTCHA solving as the happy path.

Use [choose your path](/get-started/choose-your-path) after first success when you know the capability but still need to pick SDK, MCP, REST, CDP, quick actions, or advanced exact control.