Build a Research Shortlist
Find sources around a topic before asking a model to compare or summarize them.
Firecrawl Search APION SANDBASEFirecrawl Search API
Turn a research question into web sources your Agent can inspect. Explore illustrative queries and responses below, then confirm request fields and access details in the live Playground before integrating.
Research questionRelevant sourcesAgent context
SEARCH USE CASES
Start with a Question, Not a URL List.
Discover Product & Market Context
Find documentation, product pages, and public information for a research workflow.
Ground an Agent’s Next Step
Give an Agent candidate URLs to inspect, select, and use as sources.
TECHNICAL INTEGRATION
Your Query. Your SandBase Key.
The real request path has not been verified and is intentionally omitted. Replace the request URL placeholder only after checking the live Schema / Playground. No request runs on this page.
Refer to the live Schema / Playground. Keep your key server-side. Request fields below are illustrative; confirm supported options before production.
curl -X POST '<REQUEST_URL_FROM_LIVE_SCHEMA_OR_PLAYGROUND>' \
-H "Authorization: Bearer $SANDBASE_API_KEY" \
-H 'Content-Type: application/json' \
-d '{
"query": "agent sandbox security best practices",
"limit": 3
}'Request Schema
APPLICATION/JSON| Field | Type | Required | Description |
|---|---|---|---|
query | string | Required | The search question or keywords. Documented by the SandBase Search endpoint. |
limit | integer | Optional | Maximum result count. Confirm current range and default in Playground/docs. |
READ THE RESPONSE
Sources You Can Inspect and Use.
The response below is mock data based on the public Playground envelope. It is not a live search or a guarantee of returned fields.
MOCK RESPONSE · NOT LIVE DATA
{
"data": {
"web": [
{
"title": "A practical guide to Agent sandboxes",
"url": "https://example.com/guides/agent-sandboxes",
"description": "Isolation, scoped access, and disposable environments for Agent work."
},
{
"title": "Designing scoped tool access",
"url": "https://example.com/guides/tool-access",
"description": "A checklist for controlling how Agents use tools."
}
]
}
}data.web[]- The public Playground example wraps web results in data.web; the README shows a trimmed web[] form. Confirm the envelope in your run.
title / url- A readable source label and its destination. Keep the URL for citations.
description- A source preview; it can vary in length or be absent. It is not a guarantee of a full page transcript.
position- Shown in the public example as source order. Optional fields are omitted from this mock response.
THREE RESEARCH PATTERNS
Discover First. Extract When You Need To.
Search Only
Query → candidate URLsStart with page titles, URLs, and descriptions. Select sources before collecting full content.
Search + Content
Query → sources + contentIllustrates discovery and extraction together. Confirm this option in the current SandBase schema.
Search → Scrape
Query → select URL → ScrapeSeparate discovery from extraction when your Agent needs to choose which pages to read.
ERRORS & LIMITS
Make Uncertainty Explicit.
Exact status codes, retry behavior, and current quotas have not been verified. Confirm them in Playground/docs.
Authentication
Missing or invalid workspace key
Confirm status code and key permissions in Playground/docs.Input Validation
Missing query or unsupported options
Confirm validation errors and supported fields in Playground/docs.Rate or Balance Constraints
Too many requests or insufficient balance
Confirm limits, costs, and retry guidance in Playground/docs.Empty or Unavailable Results
No useful sources, or upstream failure
Confirm response shape and failure handling in Playground/docs.THE SANDBASE ECOSYSTEM
A Tool Is a Start. An Agent Does the Work.
Web data, reusable Skills, an Agent, and the model of your choice — connected in one workflow.
01
Firecrawl API
Discover & extract
02
Skills
Package the workflow
03
Public Agent
Put the tools to work
04
Swappable Model
Choose the reasoning
EXAMPLE SKILLS · NOT VERIFIED DEPENDENCIES
Web ScraperURL to MarkdownWebsite Monitor
These labels illustrate possible workflows, not a verified catalog or automatic monitoring service.
Bring It to Your AI Tools.
SandBase MCP connects your tools to the same ecosystem. Confirm the available tool names in the docs.
CONTINUE THE WORKFLOW
Found a Page? Choose Your Next Step.
Detailed sibling pages are coming next.
Scrape
Extract clean content from one page.
Map
Discover the URLs that make up a site.
Batch Scrape
Extract content from a known URL list.
Crawl
Follow a site and collect page content.
FAQ
Before Your First Search.
How is Search different from Scrape?
Search starts from a query and discovers pages. Scrape starts from a URL you already know and extracts its content.
Can Search also return page content?
The Search + content pattern is shown as an illustrative option. Confirm the supported payload and response format in SandBase Playground or the current docs.
Are these code samples ready to run?
No. The request path is unverified and is intentionally omitted. Payloads and responses are illustrative; refer to the live Schema / Playground for the request URL, authentication, and supported fields before production.
How many results can I request?
Confirm the current default, maximum result count, and concurrency limits in Playground/docs. No allowance is assumed here.
Which model should read the results?
The web-data step is separate from the model step. Use a model suited to your task and swap it without changing the concept of the Search workflow.
Does Search guarantee current or complete coverage?
No completeness or freshness guarantee is asserted here. Inspect returned sources and timestamps where available, and verify important claims against the original pages.