MCP server with 19 tools — file access, MySQL, web, calculator, text processing
20 tools with basic schemas and descriptions, but significant gaps in parameter documentation and error handling. Most tools have descriptions (10-150 chars, below the 194-char baseline), but many lack actionable guidance. Input schemas present but sparse, most parameters have type and description, but no enums, ranges, or validation constraints. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Error responses are generic ('Error: ...') rather than recovery-guided. Two-step write flow (write_file_prepare/confirm) is a good pattern, but not replicated elsewhere. Composition is reasonable, tools are single-purpose, but output schemas are undocumented. No pagination support for list_dir or search_in_files despite potential for large result sets.
Decode base64-encoded text.
Encode text to base64.
Evaluate a mathematical expression safely. Supports +, -, *, /, **, %, parentheses, and basic functions (sin, cos, sqrt, etc.). Returns the numeric result.
Fetch a URL and parse the response as JSON. Useful for querying REST APIs. Returns pretty-printed JSON. Handles errors gracefully.
Get the current date and time in ISO 8601 format (UTC).
Compute a hash of text using the specified algorithm (md5, sha1, sha256, sha512).
Parse and pretty-print JSON. Validates JSON syntax and returns formatted output.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). LLMs cannot distinguish safe read-only tools from destructive ones without explicit hints. All 20 tools lack this metadata.
Error responses are generic ('Error: path not allowed', 'Error: file not found') with no recovery guidance. LLMs cannot self-correct or plan next steps. Should include actionable hints like 'Try search_in_files() to locate the file' or 'Allowed paths: [...]'.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | <=2025-11-25 | v2 |
List files and directories in a given path. Returns one entry per line with type indicators.
Run a READ-ONLY SQL query (SELECT only) on the demo MySQL database. Returns results as a list of dicts. Safety: only SELECT statements allowed.
Read a text file from disk (first 5000 chars). Use for inspecting code, configs, logs. Respects sandbox: only allowed_paths are accessible.
Find all regex matches in text. Returns a list of matched strings.
Replace all regex matches in text with a replacement string.
Search for a text pattern across files in a directory. Returns matching lines with file paths. Useful for finding code references, config values, or log entries across a codebase.
Fetch a webpage and return a short extractive summary (first 1000 chars of cleaned text). Useful for quickly understanding page content without reading the full page.
Get system information: platform, hostname, Python version, and current working directory.
Join a list of strings with a delimiter.
Split text by a delimiter and return a list of parts.
Count words, characters, lines, and sentences in a text. Returns a summary dict.
Confirm and execute a pending file write. Requires the token from write_file_prepare. This is the second step of the human-in-the-loop write flow.
Prepare a file write operation (human-in-the-loop). Returns a confirmation token. The file is NOT written yet — use write_file_confirm with the token to complete. This two-step process prevents accidental file modifications.
No pagination support for list_dir or search_in_files. Large directories or search results could exceed context window. Missing limit, offset, and total_count fields in output schema.
Descriptions for get_datetime and sysinfo are under 20 chars ('Get the current date and time in ISO 8601 format (UTC).' is 60 chars, but sysinfo is only 'Get system information: platform, hostname, Python version, and current working directory.', 90 chars, acceptable). However, base64_encode/decode descriptions lack context on when to use them vs alternatives.
No output schemas documented. LLMs cannot plan downstream tool calls or extract structured data. E.g., list_dir returns 'one entry per line' but no schema for entry structure; search_in_files returns 'matching lines with file paths' but no field names or types.