Skip to main content
Version: 2.27

Cycle MCP

Experimental Feature

Cycle MCP is currently an experimental feature. Tools, resources, prompts, and agent behavior may change between Cycle releases. The feature itself may be revised or removed entirely as we learn how teams use it.

Cycle ships with an MCP (Model Context Protocol) server built into cycle-cli.exe. MCP is a widely accepted protocol that lets AI coding agents discover and call external tools through a common interface. Cycle MCP uses it to expose the tools, resources, prompts, and specialized agents needed to author, run, and debug Cycle tests.

Cycle MCP is compatible with AI coding agents that integrate with MCP servers, such as Claude Code, Cursor, and GitHub Copilot.

Once you wire your agent up to the Cycle MCP server, you can describe a test in plain language. For example, you can prompt the agent like this:

"Search the staging storefront for a product, add the top result to the cart, and verify the cart count goes from 0 to 1."

The agent will:

  • Read Cycle's step catalog to find the right step definitions
  • Open a real browser, navigate the application, and capture live element locators
  • Generate the corresponding .feature file (delegating to a specialized authoring agent)
  • Validate the file's syntax
  • Execute the test with cycle-cli and read structured error details if it fails

The agent does the typing; you review and iterate.

How to Use This​

You can get started with our setup guide: to connect the Cycle MCP server with the agent of your choice. To fully understand what is possible using Cycle MCP, take a look at the Tools and Resources: that we provide to agents. For the easiest on-ramp, the Prompts: page covers the guided, fill-in-the-blank requests the server provides, and the Agents: page explains the specialized subagents the main agent delegates to. Once you're ready to try it out, the Workflows: page provides typical patterns for writing, debugging, and exploring tests with an agent. If you come across any issues, the Troubleshooting: section lists common errors and how to recover from them.

Prerequisites​

  • Cycle installed (the installer places cycle-cli.exe, which serves as both the test runner and the MCP server)
  • A supported AI coding agent: Claude Code, Cursor, or GitHub Copilot in VS Code
  • A signed-in Cycle account (the same one you use for the Cycle Desktop client). The MCP server requires you to be signed in, and the agent authenticates through the same flow as the desktop client.

How It Fits​

The Cycle MCP server is a companion to the Cycle Desktop client and cycle-cli, not a replacement. The desktop client and CLI continue to be the primary surfaces for running tests interactively or in CI/CD pipelines. The MCP server is the path for agent-driven test authoring and exploration.

Most teams adopt it incrementally: first to draft a tricky locator or a new feature file, later to maintain regression suites alongside the desktop client.