Parallel.ai 🔬

High-accuracy web research API built for AI agents. Outperforms Perplexity/Exa on research benchmarks.

APIs

Search API – Quick web search

python3 {baseDir}/scripts/search.py "Who is the CEO of Anthropic?" --max-results 5
python3 {baseDir}/scripts/search.py "latest AI news" --json

Task API – Deep research & enrichment

# Simple question → answer
python3 {baseDir}/scripts/task.py "What was France's GDP in 2023?"

# Structured enrichment (company research)
python3 {baseDir}/scripts/task.py --enrich "company_name=Stripe,website=stripe.com" \
  --output "founding_year,employee_count,total_funding"

# Research report (markdown with citations)
python3 {baseDir}/scripts/task.py --report "Market analysis of the HVAC industry in USA"

# With authenticated sources (NEW - Jan 2026, requires browser-use.com key)
export BROWSERUSE_API_KEY="your-key"
python3 {baseDir}/scripts/task.py "Extract specs from https://nxp.com/products/K66_180"

Authenticated Sources (NEW – Jan 2026)

Task API now supports authentication-gated private data sources via MCP servers:

• Internal wikis & dashboards
• Industry databases (NXP, IEEE, etc.)
• CRM systems & subscription services
• Paywalled content

Uses browser-use.com MCP integration:

Setup

1. Get API key from browser-use.com
2. Create a profile with saved login sessions (Profile Docs)
3. Set BROWSERUSE_API_KEY env var

Usage

# CLI
export BROWSERUSE_API_KEY="your-key"
python3 {baseDir}/scripts/task.py "Extract specs from https://nxp.com/products/K66_180"

# Python SDK
task_run = client.beta.task_run.create(
    input="Extract migration guide from NXP K66 docs",
    processor="ultra",
    mcp_servers=[{
        "type": "url",
        "url": "https://api.browser-use.com/mcp",
        "name": "browseruse",
        "headers": {"Authorization": "Bearer YOUR_BROWSERUSE_KEY"}
    }],
    betas=["mcp-server-2025-07-17"]
)

# cURL
curl -X POST "https://api.parallel.ai/v1/tasks/runs" \
  -H "x-api-key: $PARALLEL_API_KEY" \
  -H "Content-Type: application/json" \
  -H "parallel-beta: mcp-server-2025-07-17" \
  --data '{
    "input": "Extract data from internal dashboard",
    "processor": "ultra",
    "mcp_servers": [{
      "type": "url",
      "url": "https://api.browser-use.com/mcp",
      "name": "browseruse",
      "headers": {"Authorization": "Bearer YOUR_KEY"}
    }]
  }'

Requirements:

• browser-use.com API key + profile with saved credentials
• parallel-beta: mcp-server-2025-07-17 header
• Processor: ultra (supports multiple tool calls)
• Max 10 MCP servers per request

Capabilities:

• Navigate SPAs and JS-heavy sites
• Fill forms, click buttons, multi-step workflows
• Combine with Parallel’s public web research

Processors

Response Format

Search API

• search_id – unique identifier
• results[] – url, title, excerpts, publish_date
• usage – API usage stats

Task API

• run_id – task identifier
• output.content – structured result or markdown report
• output.basis[] – citations, reasoning, confidence per field

When to Use

• Search API: Quick lookups, current events, simple queries
• Task API:
  • Enrichment at scale (CRM, company lists)
  • Deep research reports with citations
  • Accessing authenticated/gated sources
  • Structured output with confidence levels

API Reference

Docs: https://docs.parallel.ai Platform: https://platform.parallel.ai