Loading ToolBench
MCP server that exposes a Memex wiki to Claude Code and other MCP-compatible agents. Provides tools for searching, reading, writing, and analyzing wiki pages with hybrid search, confidence tracking, and knowledge graph visualization.
Server has 11 tools with complete schemas and descriptions. Naming follows verb_noun convention (wiki_search, wiki_read, wiki_write). Descriptions are present and adequate (avg ~120 chars), meeting the 10-1024 char baseline. However, several critical gaps: (1) Output schemas are NOT documented, no tool specifies what fields are returned, violating pattern:tool and mxe:strip-api-responses. (2) Error handling is absent, no guidance on recovery, retryability, or actionable error messages. (3) Parameter descriptions lack constraints (e.g., wiki_ingest 'source_path' has no format/validation hints; wiki_write 'slug' lacks regex pattern). (4) No pagination support visible despite tools like wiki_search and wiki_list that likely return multiple results. (5) Tool composition issues: wiki_read and wiki_read_by_slug overlap; wiki_query appears to synthesize answers but no output schema shown. (6) No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels (WRITE vs READ_ONLY). These gaps place the server in the C/D range despite complete schemas.
Analyze wiki confidence: distribution, stale pages, contradictions.
Get the wiki knowledge graph as nodes and edges for visualization.
Ingest a new source document into the wiki. Creates summary and updates relevant pages.
Check wiki health: broken links, missing citations, orphan pages, contradictions.
List all pages in the wiki, optionally filtered by type (sources, entities, concepts, synthesis).
Ask a natural language question. Searches wiki and synthesizes an answer with citations.
Read a wiki page by path (e.g., 'concepts/zero-rag' or 'entities/karpathy').
Output schemas not documented. No tool specifies what fields are returned (e.g., wiki_search returns 'matching pages with snippets' but structure unknown). LLMs cannot plan downstream calls or extract data without knowing response shape.
No error handling guidance. Tools lack recovery hints, retryability classification, or actionable error messages. E.g., wiki_read_by_slug mentions 'ambiguous slug' error but no guidance on how LLM should respond.
Parameter constraints missing. 'source_path' in wiki_ingest lacks format/validation hints; 'slug' in wiki_write has no regex pattern; 'query' in wiki_search has no length bounds. LLMs cannot validate inputs before calling.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
Read a wiki page by bare slug (no type prefix), searching entities/concepts/sources/synthesis for a match. Returns found=false if missing, or an error listing candidates if the slug is ambiguous across type dirs.
Search the Memex wiki using keywords. Returns matching pages with snippets.
Get wiki statistics: page counts, link density, health metrics.
Write a new wiki page (or overwrite existing). Creates file at wiki/{type}/{slug}.md and commits with agent attribution.
No pagination support visible. wiki_search and wiki_list accept 'limit' but no 'offset', 'cursor', or 'page' parameter. No 'total_count' or 'next_cursor' in response schema. Large result sets will blow context window.
Tool annotations missing. No readOnlyHint, destructiveHint, or idempotentHint despite clear risk levels (WRITE vs READ_ONLY declared in metadata). Agents cannot distinguish safe from destructive operations.