CapSolver Agent Automation: MCP, Agent Tools, and Core SDK
CapSolver Agent Automation is now live as a technical integration layer for browser-based AI agents.
A browser agent may complete most of a task correctly and still stop when a CAPTCHA appears. This release gives the existing runtime a defined capability to call at that point, instead of placing CAPTCHA logic inside the prompt or rebuilding it for every framework.
The release provides three ways to add CAPTCHA handling to an existing workflow:
capsolver-core— the Python engine for direct browser and Playwright automation.capsolver-agent— tool schemas and an executor for agent frameworks.capsolver-mcp— an MCP server for compatible AI clients.
Integration paths are available for MCP, LangChain, Browser Use, OpenAI Agents SDK, and Playwright.
The runtime flow stays simple:
agent or browser → CapSolver tool call → capsolver-core → structured result → original workflow
For MCP clients, capsolver-mcp exposes five tools: solve_captcha, detect_captchas, solve_on_page, get_balance, and get_supported_captchas. Agent frameworks can use capsolver-agent, while teams controlling Playwright directly can call capsolver-core without adding an LLM layer.
Your existing stack continues to manage the agent, browser session, permissions, and final task. CapSolver handles the supported CAPTCHA step and returns the result to the same workflow.
Links
GitHub
CapSolver is intended for lawful, user-authorized automation workflows. Feedback on tool schemas, MCP setup, errors, and framework integrations is welcome.