Loading ToolBench
Your AI's private doorway into your Second Brain (MCP Server)
NeuroNest exhibits significant gaps in definition quality. While all 8 tools have descriptions and mostly clear naming, the implementation suffers from: (1) incomplete input schemas, most parameters lack explicit type definitions and descriptions in the visible code; (2) no documented output schemas, return types are inferred from docstrings, not formally specified; (3) missing parameter constraints and validation guidance; (4) weak error handling and recovery instructions. Tool names follow verb-noun convention (estimate_, append_, get_, search_, read_), which is positive, but parameter annotations are sparse. The semantic search tool (search_notes) has the most complete schema visible, while token estimation and graph tools have minimal parameter documentation. This is a typical community server, functional but below production quality.
Appends text to today's daily note. Creates the note if it doesn't exist. The daily note is stored in a 'Daily' folder in the root of the default vault, formatted as YYYY-MM-DD.md.
Scans all configured vaults for notes modified in the last 24 hours. Extracts TODOs, tasks, and new headings to build an executive summary.
Estimates the number of tokens in the given text using standard LLM tokenization (tiktoken cl100k_base).
Finds all notes that link to the specified note using wiki-style link syntax [[note_name]].
Finds notes related to the specified note by outbound links and shared tags.
Reads a markdown note from the user's vault. filename: relative path to the markdown file (e.g., 'ideas/project.md') heading_filter: return only this heading's content token_cap: limit the returned tokens (default 4000)
Missing output schemas: No tools document their return types formally. LLMs cannot plan downstream calls or extract required fields (e.g., note_id, file_path) without seeing return structure.
Parameter descriptions incomplete or missing. 'hours' param in recent_changes has description but no type constraint (e.g., min=1, max=365). 'heading_filter' and 'token_cap' in read_note lack validation guidance.
Error handling provides no recovery guidance. Tools fail silently or return generic strings. No indication of whether an error is retryable, user-fixable, or fatal. E.g., read_note with invalid filename should suggest 'Use search_notes to find the correct path.'
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Returns a list of files modified in the last `hours` hours across all configured vaults, extracting the most recently modified sections (or the end of the file/highlights).
Searches all markdown notes in the configuration for the given query. Uses hybrid search (keyword + semantic similarity).
No pagination or result limits declared. search_notes accepts max_results but recent_changes and daily_briefing do not. Tools returning lists (get_backlinks, get_related_notes, recent_changes) should document expected cardinality and pagination strategy.
Destructive operation (append_to_daily_note) lacks confirmation/dry-run. WRITE operations should support a dry_run parameter or return a confirmation request before executing. Currently, agent can append unwanted text without preview.
Parameter type specificity weak. 'filename' in read_note should clarify format: 'Relative path to vault root (e.g., ideas/project.md). No leading slash or absolute paths.' Missing guidance invites path traversal attempts.