High-performance MCP server for large repositories using ripgrep
This server demonstrates solid definition quality with well-structured tool names, clear descriptions, and complete input schemas. All 4 tools follow verb_noun naming conventions and include proper JSON Schema validation. Tool descriptions are adequate (100-150 chars each) and explain the core functionality. Parameter schemas are comprehensive with type constraints, minLength/maxLength, min/max bounds, and enum-like behavior through defaults. However, there are gaps in parameter descriptions (some fields lack explanatory text for why they exist), output schemas are not explicitly documented, and error handling guidance is minimal. The code shows defensive validation (asString, asIntegerInRange) but error messages are generic without recovery hints.
List repository files via ripgrep --files with optional glob filters and bounded output.
Search the repository with ripgrep. Supports optional glob filters and bounded results.
Read a specific file line range within the repo root. Maximum 500 lines per call.
Search for an exact symbol using ripgrep word boundaries. Uses TypeScript or Python mode based on project detection.
Parameter descriptions are missing or minimal. 'globs' in project_search_rg has no description explaining that these are file path patterns to filter search scope. 'maxMatches' lacks context about when to use smaller vs larger limits. These omissions force LLMs to infer parameter meaning from names alone, violating pattern:tool-description.
Output schemas are not documented. The code returns SearchMatch objects with nested submatches and TruncateReason enums, but the tool descriptions do not explain what fields the LLM should expect (path, line, column, text, submatches, truncated). LLMs cannot plan downstream calls or extract chaining IDs without this. Violates pattern:tool and mxe:include-chaining-ids.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 54 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Error handling lacks recovery guidance. Code throws generic errors ('Invalid input', 'must be a string') without actionable next steps. Example: when read_range fails due to invalid line numbers, should suggest calling list_files to enumerate available files, or project_search_rg to find a symbol in a file first. Violates pattern:recovery-guide.
Tool descriptions do not explain WHEN to use each tool or how they differ. 'project_search_rg' and 'symbol_search' both search the repo but with different semantics (regex vs word boundary). An LLM cannot decide which to call without explicit guidance. Violates pattern:tool-description.
No pagination support or result limits documented. list_files maxResults caps at 500, but description does not explain what happens when results exceed the limit (truncated? next_cursor?). project_search_rg maxMatches defaults to 100 but does not document if results are stable/deterministic across repeated calls. Violates pattern:paginated-result.
Parameter 'globs' accepts up to 50 glob patterns but the description does not explain the expected format (e.g., '*.ts', '**/*.test.js'). LLMs may pass invalid globs or overly broad patterns. Violates review:param-validation-rules.