| Integration | Best for | Auth model | Docs |
|---|---|---|---|
| CLI + Skills | Coding agents with terminal access | API key or OAuth login | CLI, Agent Skills |
| MCP Servers | Chat assistants and LLM-powered apps | OAuth or API key | MCP Quickstart, Programmatic Use |
| SDK / Native Tool Calling | Production agents needing full control | API key | Python, TypeScript |
Decision guide
- I use a coding agent
- I use a chat assistant
- I'm building my own app
Use the CLI with a plugin or Agent Skills.If your agent has terminal access — Claude Code, Cursor, Windsurf, Cline, GitHub Copilot — the CLI is the most capable integration. It exposes all Parallel APIs (search, extract, research, enrich, findall, monitor) and produces structured JSON that agents parse natively.CLI commands are also deeply embedded in LLM training data. Models already know how to compose shell commands, pipe output, and handle JSON — no schema overhead required.Why not MCP? In terminal-based agents, CLI tools consume far fewer context tokens than MCP tool schemas. An MCP server dumps its full tool catalog into the context window upfront. A CLI call like
parallel-cli search "query" --json costs only the tokens for the command and its output.| Agent | Recommended setup |
|---|---|
| Claude Code | Claude Code Plugin (installs CLI + slash commands + auto-skills) |
| Cursor | Cursor Plugin or Agent Skills |
| Cline, Copilot, Windsurf | Agent Skills (uses CLI under the hood) |
| Other CLI agents | Install the CLI directly and use --json output |
Quick comparison
API coverage
| Capability | CLI | MCP | SDK |
|---|---|---|---|
| Search | Yes | Yes (Search MCP) | Yes |
| Extract | Yes | Yes (Search MCP web_fetch) | Yes |
| Deep Research | Yes | Yes (Task MCP) | Yes |
| Data Enrichment | Yes | Yes (Task MCP) | Yes |
| FindAll | Yes | No | Yes |
| Monitor | Yes | No | Yes |
Authentication
| Method | How it works |
|---|---|
| CLI | parallel-cli login opens a browser for OAuth, or parallel-cli login --device for headless/SSH environments. You can also set PARALLEL_API_KEY as an environment variable. Credentials are stored locally. |
| MCP | In chat assistants, OAuth is built in — your client handles the browser-based auth flow automatically. For programmatic use, pass your PARALLEL_API_KEY as a Bearer token. |
| SDK | Set PARALLEL_API_KEY in your environment. For multi-user apps, use the OAuth Provider to obtain keys on behalf of users. |
Get started
CLI + Skills
Install the CLI, authenticate, and add agent skills for your coding environment.
MCP Servers
Connect Search MCP or Task MCP to a chat assistant or your own app.