Parallel CLI

One interface for Parallel web data work: extract, deep research, enrichment, findall, memory, and monitor. CLI-backed workflows require parallel-cli installed, authenticated, and funded; the search routing below avoids the CLI where possible.

Routing: MCP first, CLI for depth

General web search, quick lookups, fact-checks: use the parallel-search MCP server (web_search) when it is configured. It is free. Do not call parallel-cli search for these. For current official facts (framework/library/API documentation), prefer the harness's native web search per repo convention. When the MCP server is absent, use the harness's native web search, or fall back to parallel-cli search.
Escalate to parallel-cli search only when MCP is insufficient: --mode advanced for hard multi-step questions, --after-date freshness, --include-domains / --exclude-domains, --location geo-targeting, or high-volume batch queries.
URL content extraction: parallel-cli extract (cleaner than MCP web_fetch for PDFs, JS-heavy pages, and long articles).
Deep/exhaustive research (user says "deep", "exhaustive", "comprehensive report", "thorough investigation"): prefer the repository's deep-researcher skill for deep cited research when available (repo convention); research run + poll is the engine and the route when deep-researcher is unavailable.
Structured entity lists ("find all X", "list every Y"): findall. Add fields to a list you already have: enrich.
Recall prior runs: memory. Continuous change tracking: monitor.
Do not use the parallel-task MCP for research or enrichment; the CLI is the tool of record.

Setup

Check and install, requiring parallel-cli >= 0.8.1: