Parallel provides web infrastructure APIs for AI agents: searching the web, extracting page content, running multi-step research, monitoring for changes, and finding entities that match a description. Each API is designed for a model to call rather than a person to browse, so results come back as LLM-ready excerpts or structured JSON with citations.
The Search API takes a natural-language objective plus a few keyword queries and returns compressed excerpts in one round trip. The Extract API turns up to 20 URLs per request into clean markdown, including JavaScript-rendered pages and PDFs. The Task API accepts a plain-language request or a JSON schema and performs multi-hop research, returning per-field citations in an output.basis block that records the sources and reasoning behind each value.
Task processors are tiered from lite through base, core, pro, ultra, and ultra8x, trading latency for depth and field count, with -fast variants for lower latency; Search offers basic, turbo, and advanced modes. Access is by API key from the platform console, with SDKs for Python and TypeScript and a remote MCP server that exposes Search to coding agents.
Features
- Search API: objective-driven web search returning LLM-optimized excerpts with citations
- Extract API: URL-to-markdown conversion for up to 20 URLs per request, with optional content targeting
- Task API: multi-hop research and enrichment against a schema, running from seconds to hours by processor tier
- Monitor API: scheduled queries at intervals from one hour to 30 days, with webhook alerts on material change
- FindAll (beta): discovers and verifies companies or people matching natural-language criteria
- Entity Search: synchronous people and company lookup for latency-sensitive workflows
- Provenance: per-field citations and reasoning in output.basis for every task result
- MCP server: remote Search MCP for Claude Code, Codex, Cursor, and VS Code
- SDKs: Python on PyPI and TypeScript on npm
