This Laravel helper MCP server has 3 tools with partial schemas and descriptions. Tool naming is clear and action-oriented (tail_log_file, search_log_errors, run_artisan_command), following verb_noun convention. However, descriptions are present but minimal (10-50 chars), and critical details are missing: no return type documentation, no output schema specification, no error guidance, and no security annotations. The run_artisan_command tool is particularly dangerous, it accepts arbitrary shell commands with no validation, no destructive hint annotation, and no confirmation pattern. Input parameters have type information but descriptions are sparse. The code shows file I/O and subprocess execution, but the tool definitions themselves lack the metadata needed for LLMs to reason about side effects and recovery paths.
Run an artisan command in the Laravel directory
Search the Laravel log file for errors within a specified time window
Tail the last N lines of the Laravel log file
Critical: run_artisan_command accepts arbitrary strings with no validation or command allowlisting. No destructiveHint annotation. Accepts any artisan command (dangerous for commands like db:wipe, migrate:reset) without confirmation or dry-run support.
No output schema documentation. Tools return strings (raw stdout) with no structured field definitions. LLMs cannot extract or chain results, tail_log_file returns unparsed log lines, search_log_errors returns raw grep output. Violates pattern:response-shaper.
No error guidance. Error handling returns raw messages ('Error: Laravel directory not initialized') with no actionable recovery steps. Does not tell LLM what to do next (e.g., 'Set LARAVEL_PATH env var and restart').
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-21 | F | 49 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 51 | - | v1 |
Tool descriptions are below 50 characters and lack context. 'Tail the last N lines of the Laravel log file' (51 chars) omits WHEN to use it, WHAT to expect, or any prerequisites. LLMs cannot reliably select this tool in context-heavy scenarios.
No security annotations. run_artisan_command is destructive (can execute migrations, deletes, etc.) but has no readOnlyHint or destructiveHint. search_log_errors reads stateful tracking from disk; no idempotentHint. LLMs have no way to reason about side effects.
Parameter descriptions are minimal. 'lines' param described as 'Number of lines to tail from the log file' with no mention of range (10 is default but what's max?). 'minutes_back' says 'max: 60' in description, not in schema constraints. 'command' for run_artisan_command has no format guidance, examples of allowed commands, or warnings about destructive commands.
No idempotency guarantees. search_log_errors tracks state on disk ('show_all' flag, 'last_file_check'). Retries may behave differently depending on file state. Pattern:idempotent-operation violated.
Input schema incomplete. 'minutes_back' constraint (1-60) is documented in description but NOT in JSON Schema minValue/maxValue. 'lines' has a default but no min/max bounds. JSON Schema constraints are machine-readable; description-only constraints are not.
Hard dependency on environment variable (LARAVEL_PATH). If not set, server fails at initialization. No fallback, no graceful degradation. Users may not know why the server is unresponsive.