Model Context Protocol (MCP) server for Cypress, enabling AI-powered test automation in Cursor
This MCP server provides 8 tools for Cypress test automation with explicit tool registration and schemas visible in src/index.js. Naming follows verb_noun patterns (cypress_run, cypress_open, cypress_validate). Descriptions are present for all tools (ranging 60-180 chars, within baseline 34-392 range). However, critical gaps exist: parameter descriptions are sparse or missing entirely (many params lack actionable guidance), output schemas are completely undocumented (no tool returns a documented structure showing what fields the agent should expect), and error handling lacks recovery guidance. The cypress_validate tool's schema has a logical flaw (requires 'testFile' but describes 'testCode' as alternative, forcing LLMs to guess). No tool documents whether it is idempotent, retryable, or destructive, critical for agent safety. Parameter validation rules are missing (e.g., no mention of valid enum values for many params, no range/format constraints). Overall, tools are functional but underspecified for LLM-based agents relying on descriptions to select and invoke them correctly.
Execute a Cypress command with chaining support. This is a lower-level tool for executing specific Cypress commands.
Generate a Cypress test file from a description
Get test results from a previous Cypress run by run ID, or all results if no ID provided
Get screenshots captured during test runs
Get videos recorded during test runs
Open Cypress Test Runner in headed mode for interactive debugging
Run Cypress tests with specified options. Returns test results including pass/fail counts, duration, and run ID.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract chaining IDs (run_id, test_path, etc.) if response structure is unknown. This violates pattern:response-shaper and makes agent composition fragile.
Parameter descriptions are missing or generic. Example: 'env' in cypress_run has description 'Environment variables to pass to Cypress' but does not specify format (object of key-value pairs? JSON string?), expected keys, or examples. 'args' in cypress_execute_command lacks any description of what values it accepts.
cypress_validate schema contradicts itself: 'required: ["testFile"]' but description says '(alternative to testFile)' for testCode. LLMs cannot determine if both are required, one is required, or one-of-two logic applies.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 52 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 42 | - | v1 |
Validate a Cypress test file or test code for syntax and best practices
No error handling guidance. Tool descriptions do not indicate what errors are retryable (transient Cypress failures) vs fatal (invalid test file path). Agents cannot distinguish 'retry this' from 'report failure to user'.
Destructive operations lack confirmation pattern. cypress_run and cypress_generate modify test state/files, but tool descriptions do not mention dry-run, confirmation, or rollback capability. Agents may accidentally overwrite tests.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). LLMs cannot infer side effects: is cypress_open safe to call multiple times? Does cypress_run persist state that affects retry behavior?
Enum constraints undocumented in descriptions. 'browser' parameter lists enums in schema but descriptions do not explain what each browser implies (headless support, platform availability, performance). LLMs may pick wrong browser without context.
No pagination support documented. Tools like cypress_get_screenshots, cypress_get_videos, and cypress_get_results likely return large datasets but lack limit, offset, or cursor parameters. No documentation of result caps (will 10,000 screenshots be returned or truncated at 50?).