Loading ToolBench
MCP server that protects your AI context window via sandboxed execution and smart indexing
Logica Context provides 17 tools with consistent naming (lctx_ prefix) and reasonable descriptions (avg ~120 chars). All tools have input schemas with type definitions. However, parameter descriptions are sparse or missing entirely, most params lack guidance on format, constraints, or valid ranges. Output schemas are undocumented. Error handling is minimal; no recovery guidance or actionable error messages visible. Tool composition is reasonable (each does one thing), but several tools combine read+write concerns (e.g., lctx_batch_execute runs commands AND indexes results). STDIO transport caps protocol readiness at 50.
Run multiple shell commands and search queries in a single call. Results are auto-indexed into the knowledge base. Returns a concise summary.
Show context budget: tokens consumed, tokens saved, percentage used, warnings.
Diagnose the Logica Context installation and check dependencies.
Execute code in a sandboxed environment. Supported languages: shell, node, python, ruby, php, go, swift, rust, deno. Output is auto-indexed.
Process a file by running code in its directory context. Output is auto-indexed.
Fetch a URL, convert HTML to markdown, and index the content. Cached for 24 hours.
Parameter descriptions are minimal or absent. 'language' param in lctx_execute lists supported values in description but lacks format guidance. 'limit' params lack min/max bounds. 'ttl_hours' in lctx_index has no constraint documentation (0 = permanent, but what's the max?).
Output schemas are completely undocumented. No tool declares what fields it returns, what types they are, or how downstream tools should consume them. LLMs cannot plan multi-step chains without knowing response structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Index git status: branch, diff, recent commits, modified files. AI starts knowing what changed.
Index text content into the knowledge base for later search. Chunks large content automatically.
List active MCP servers and their estimated context impact. Shows recommendations for optimization.
Clear all entries from the knowledge base.
Scan project structure and create a DNA profile (stack, languages, frameworks, patterns). Auto-indexes the result.
Search the knowledge base using BM25 full-text search with Porter stemming. Supports multiple queries.
Semantic search using embeddings (Voyage AI or OpenAI). Finds content by meaning, not just keywords. Requires VOYAGE_API_KEY or OPENAI_API_KEY.
Show knowledge base statistics and context savings info.
Push content to the team knowledge base (shared via Supabase). Other developers can search it.
Search the team knowledge base. Finds content indexed by any team member.
Check for and install updates to Logica Context.
No error handling guidance. Tools like lctx_execute (which runs arbitrary code) and lctx_purge (destructive) have no documented error cases, recovery paths, or confirmation mechanisms. Agents cannot distinguish retryable from fatal failures.
Tool naming combines multiple concerns: lctx_fetch_and_index (fetch + index), lctx_batch_execute (execute + index). Per pattern:tool, each tool should do one thing. Split into separate tools so agents can compose them independently.
No pagination or result limits documented. lctx_search and lctx_semantic accept 'limit' but do not specify min/max or default behavior when limit is omitted. Large result sets could exhaust context window.