Skyvern's MCP integration enables plugging any MCP-compliant large language model into Skyvern's browser automation platform. By adopting MCP, Skyvern becomes vendor-agnostic, allowing an LLM to reason, plan, and issue browser actions through the same interface, with vision-based reasoning support for mapping UI elements to actions. The MCP server supports a broad ecosystem of providers, including OpenAI, Anthropic, Azure OpenAI, AWS Bedrock, Gemini, Ollama, OpenRouter, and other OpenAI-compatible endpoints. Users select and switch models via standard MCP mechanisms, using LLM_KEY and SECONDARY_LLM_KEY environment variables and optionally constrain output length with LLM_CONFIG_MAX_TOKENS. Outputs can be structured with data_extraction_schema, enabling consistent JSON-like results from MCP-driven runs. When combined with Skyvern's Tasks and Workflows, MCP enables scalable automation across many websites, including form filling, data extraction, file downloading, and other browser actions. Full MCP documentation is available in the repository's MCP docs.