Warm-process MCP server for PHPUnit. Keeps the test framework bootstrapped across calls — faster per-suite invocation vs cold CLI. Compatible with any MCP client (Claude Desktop, Cline, Continue, Zed, custom).
Single tool 'phpunit_run' has a clear action verb and reasonable description (89 chars). Input schema is present with three optional parameters, all typed as strings with descriptions. However, output schema is not formally documented in the tool definition, only inferred from the return type hint in the PHP docblock. Parameter descriptions lack format constraints (e.g., what constitutes a valid --filter pattern). No enum constraints despite filter/group accepting specific PHPUnit syntax. Error handling returns structured objects with error_class and trace, which is good, but descriptions do not guide LLM recovery (e.g., 'SecurityError' is returned but not explained in the tool description). Security containment logic is present and tested, but not surfaced in the tool description as a constraint. Overall: functional but lacks LLM-optimized descriptions and formal output schema documentation.
Run PHPUnit tests. Server-pinned config. Returns JUnit XML output + exit code.
Output schema not formally documented. Return type is inferred from PHP docblock (array with exit_code, output, warm_boot, error, error_class, trace) but not exposed as a structured JSON Schema in the tool definition. LLMs cannot reliably parse or chain on undocumented output fields.
Parameter descriptions lack actionable format constraints. 'filter' and 'group' accept PHPUnit-specific syntax (e.g., --filter 'ClassName::methodName') but descriptions do not explain valid patterns. LLMs will guess and pass invalid values.
Security containment (path traversal guard) is implemented but not documented in the tool description. LLMs do not know that testFile is restricted to the working directory, they may attempt to pass absolute paths outside the configured scope and receive cryptic SecurityError responses.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2025-06-18+ | v2 |
Error responses (SecurityError, exception traces) are returned but the tool description does not guide LLM recovery. A SecurityError should suggest 'Ensure testFile is within the configured working directory', currently the LLM must infer this from the error message alone.
STDIO transport only. Server is not remotely accessible and cannot be used by hosted MCP clients (e.g., Claude.ai, Cline cloud). Limits deployment to local/desktop environments.