Agents

AI coding agents — Claude Code, OpenAI Codex, Cursor, Cline, and others — frequently hit captcha walls while browsing, scraping, or automating the web. NoCaptchaAI lets you hand that capability to the agent so it can clear the challenge and keep going.

There are two ways to wire it up:

Call the API directly. Any agent that can make HTTP requests can POST /createTask and poll POST /getTaskResult against https://api.nocaptchaai.com. This works everywhere but the agent has to know the request shape each time.
Use an MCP server (recommended). A Model Context Protocol server wraps those endpoints as first-class tools (create_task, get_task_result, solve, balance). The agent then sees "solve this captcha" as a native action, with typed inputs and outputs, instead of hand-rolling HTTP calls.

Every integration below builds on one of these two foundations. Start with the MCP page — it is the source of truth for the server config the agent pages reuse.

Model Context Protocol (MCP)

The foundational server that exposes solving as agent tools. Read this first.

Claude Code

Add the MCP server with claude mcp add, then ask Claude Code to solve.

OpenAI Codex

Cursor

Add the MCP server via .cursor/mcp.json and use it in Composer.

Cline

Add the MCP server in Cline's MCP settings.

OpenClaw

Wire up MCP or call the API directly from an autonomous browser agent.

More integrations on the way

This list keeps growing. Every agent we support — and any agent you bring yourself — builds on the same MCP server or the raw API. Once you understand those two, adding a new agent is just a config snippet.

Grab an API key from the dashboard before you start.