Loading ToolBench
MCP server for semantex semantic code search — hybrid dense + BM25 retrieval engine with support for 25+ programming languages
Semantex MCP provides 6 well-named tools with clear action verbs (semantex_agent, semantex_search, semantex_deep, semantex_index, semantex_memory_save, semantex_memory_recall). All tools have descriptions (194 - 250 chars, within baseline 34 - 392 range). Input schemas are present with type declarations and parameter descriptions. However, output schemas are not documented, LLMs cannot predict response structure for downstream chaining. Parameter descriptions lack format constraints (e.g., 'depth' accepts 'quick|search|deep' but no enum declared). Error handling guidance is absent, tools do not indicate what to do on failure. Memory tools (save/recall) lack scope validation examples. Overall: solid naming and descriptions, but incomplete schemas and missing error recovery patterns limit production readiness.
Intelligent code search with automatic query classification. Analyzes the query, picks the best strategy (semantic, exact symbol, graph walk, deep search, regex, file pattern), executes with fallbacks, and returns a pre-formatted answer. Default for all code search queries.
One call replaces 5-10 grep+read iterations. Searches, reads, graph-expands and summarises into a prose answer with sources. Prefer `semantex_agent` for most queries; use this only when you need structured JSON.
Build or update the semantex search index. Usually NOT needed — semantex auto-indexes on first search. Call only to force a rebuild after major changes.
Recall notes from the project memory store. Search by query, scope, or retrieve all notes. Returns the most relevant matches ranked by relevance.
Save a note to the project memory store. Notes are scoped (global, file, module, etc.) and tagged for later recall. Useful for recording findings, decisions, or context that should persist across multiple searches.
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination) for downstream tool chaining. semantex_agent, semantex_search, semantex_deep return ranked results but response format is opaque.
Enum constraints not declared in schema. 'depth' parameter accepts 'quick|search|deep' but is typed as string with no enum. 'focus' accepts 'implementation|callers|signatures|patterns' but no enum. LLMs may hallucinate invalid values.
No error handling guidance. Tools do not document failure modes (e.g., 'path not found', 'index corrupted', 'memory store full') or recovery steps. Agents cannot self-correct on failure.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | <=2025-11-25 | v2 |
Find code by meaning or exact match (`grep_mode=true`). Returns ranked file chunks with paths, lines and scores. 25+ languages supported. Prefer `semantex_agent` for most queries; use this only when you need structured JSON.
Memory scope parameter lacks validation examples. 'scope' in semantex_memory_save accepts 'global|file:...|module:...' but format is not formally specified. No examples of valid scope strings provided.
Pagination not documented for semantex_search and semantex_memory_recall. 'max_results' and 'limit' parameters exist but no indication of total count, next_cursor, or whether results are truncated. Large result sets may blow context window.