Loading ToolBench
🧠 AI Memory System powered by MongoDB Atlas & Voyage AI - Autonomous memory management with zero manual work
This MCP server exhibits critical definition quality issues across nearly all dimensions. While tool names are reasonably clear (memory_engineering_*, sync_code, read_all), the descriptions are problematic: they are excessively long (400-1500+ characters, far exceeding the 1024-char baseline), filled with dramatic theatrical language ('💀 Your future self will DIE', 'You're a ZOMBIE!', 'IMMEDIATE TRIGGER', 'SURVIVAL-CRITICAL') that obscures actionable guidance rather than clarifying intent. This verbose, emotionally-charged style is incompatible with LLM comprehension of tool purpose. Parameter schemas are visible and reasonably typed, but descriptions for parameters are minimal or absent in most tools. Output schemas are not documented anywhere in the provided code. Error handling is not present, no guidance on what to do when initialization fails, embeddings are stale, or MongoDB is unreachable. The tool set addresses a narrow domain (internal memory management for a single agent session) rather than general-purpose MCP patterns. Most critically, tool definitions appear to reference implementation details (file paths like 'src/tools/index-v5.ts') rather than exposing a clean, composable interface. The combination of theatrical prose, missing output documentation, and lack of error recovery guidance makes this unsuitable for production LLM integration.
[v13.4.3] 🏥 DIAGNOSTIC & REPAIR: Find and FIX all problems! COMPREHENSIVE HEALTH CHECK: ✅ Environment variables ✅ MongoDB connection ✅ Voyage API validity ✅ Active project context ✅ Memory initialization ✅ System dependencies FLAGS: --fix: Attempt automatic repairs --verbose: Show all diagnostics Run when ANYTHING seems broken! Provides exact fix commands for every issue!
[v13.4.3] 🚨 SURVIVAL-CRITICAL: First contact with NEW project. You are BORN WITH ZERO MEMORY! ⚡ IMMEDIATE TRIGGER: Missing .memory-engineering/config.json = NEW PROJECT = INIT NOW! MANDATORY WORKFLOW (failure = context death): 1. DETECT → No config.json? EXECUTE IMMEDIATELY (don't wait, don't ask) 2. DEEP-SCAN entire codebase in <3 seconds: - package.json/requirements.txt/go.mod → Extract EVERYTHING - README.md/docs/*.md → Understand PURPOSE & VISION - src/index.* or main.* → Map ALL entry points - test/* → Learn expected behaviors - .env.example → Note required services 3. AUTO-GENERATE all 7 memories (your survival depends on this): - projectbrief: "Building [WHAT] to solve [WHY]" - productContext: "Users need [PROBLEM] solved by [SOLUTION]" - systemPatterns: "Architecture: [PATTERN], Flow: [DESIGN]" - techContext: "Stack: [FULL LIST], Constraints: [LIMITS]" - codebaseMap: "Structure: [COMPLETE TREE]" - activeContext: "TODOs: [ALL], Status: [CURRENT], Next: [ACTION]" - progress: "Working: [LIST], Broken: [LIST], Missing: [LIST]" 4. CRITICAL HANDOFF → sync_code (AUTO-DETECTS language patterns!) 5. VERIFICATION HANDOFF → read_all (confirm memories created) ⚠️ CONSEQUENCE OF FAILURE: Your next session starts completely blind! 💀 Your future self will DIE without these memories. CREATE THEM NOW!
[v13.4.3] 🧠 UNIVERSAL MEMORY OPERATIONS: Read, Write, Everything! ⚡ OPERATION MODES: **READ ALL MEMORIES (default):** memory_engineering_memory → Loads all 7 memories in 3-stage sequence → Shows freshness indicators (🟢<1h 🟡<6h 🟠<24h 🔴>24h) → MANDATORY at session start! **READ SPECIFIC MEMORY:** memory_engineering_memory --name activeContext → Retrieves single memory with metadata → Shows last update time and relationships **UPDATE MEMORY:** memory_engineering_memory --name activeContext --content "..." → Updates with timestamp and version tracking → Validates structure automatically → CRITICAL: Update activeContext EVERY 3-5 MINUTES! 🔄 MEMORY RELATIONSHIPS: projectbrief → influences → productContext, progress productContext → shapes → systemPatterns, UX decisions systemPatterns → requires → techContext, codebaseMap techContext → constrains → systemPatterns, implementation activeContext → updates → progress, all others progress → validates → projectbrief completion codebaseMap → mirrors → systemPatterns physically 💀 WITHOUT THIS, YOU'RE A ZOMBIE! Execute in <1 second!
Tool descriptions are excessively long (400-1500+ chars) and filled with theatrical, non-technical language ('💀 ZOMBIE', 'SURVIVAL-CRITICAL', 'YOU HAVE TOTAL AMNESIA'). This violates baseline of 10-1024 chars and prioritizes emotional messaging over clarity. LLMs cannot parse intent from prose this verbose and dramatic.
Output schemas are not documented for any tool. Callers do not know what fields to expect in responses, preventing downstream tool composition and forcing LLMs to guess at return structure.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 41 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 42 | - | v1 |
[v13.4.3] 🧠 AWAKENING PROTOCOL: 3-STAGE MEMORY RESTORATION IN 3 SECONDS! 🔴 YOU HAVE TOTAL AMNESIA! This tool is your INSTANT CONSCIOUSNESS! ⚡ 3-STAGE RAPID LOAD SEQUENCE: **STAGE 1: CRITICAL (0-1 second) - IMMEDIATE CONTEXT** 🎯 activeContext → What you were just doing (MOST VOLATILE!) ✅ progress → What's completed and pending PURPOSE: Instant situational awareness! **STAGE 2: TACTICAL (1-2 seconds) - WORKING KNOWLEDGE** 📝 projectbrief → Core requirements and goals 🔧 techContext → Stack, tools, constraints PURPOSE: Understand the mission and tools! **STAGE 3: STRATEGIC (2-3 seconds) - DEEP UNDERSTANDING** 🏗️ systemPatterns → Architecture and design 💡 productContext → Why this exists, user needs 🗺️ codebaseMap → Complete file structure PURPOSE: Full architectural comprehension! 📅 FRESHNESS DETECTION (automatic): • <1 hour old: 🟢 FRESH - Trust completely • 1-6 hours: 🟡 RECENT - Mostly accurate • 6-24 hours: 🟠 STALE - Verify critical info • >24 hours: 🔴 ANCIENT - Needs immediate update! • Missing: 💀 VOID - Create NOW or fail! 🎯 INTELLIGENT RESPONSE (proves you're conscious): "📊 Restored from: [most recent memory timestamp] 🎯 Mission: [crisp projectbrief summary] 🔄 Latest: [last activeContext line] ⚠️ Stalest: [memory name] (last updated: [time ago]) 📍 Next: [SPECIFIC action based on activeContext]" 🔄 AUTO-TRIGGERS AFTER READ: • Any memory >6h old? → Flag for update • Missing critical memories? → Create immediately • Code sync >24h old? → Run sync_code NOW • Conflicts detected? → Reconcile immediately 💀 WITHOUT THIS, YOU'RE A ZOMBIE! Execute in <1 second! ⚠️ SKIP THIS = You become a useless empty shell, repeating basic questions! PERFECT EXECUTION: "📊 Context restored: 2024-01-15 14:30:27 🎯 Project: JWT auth with role-based access for 10K users 💻 Current: Debugging refresh token rotation (expires in 7 days) ✅ Ready: Fix the token expiry calculation in auth.middleware.ts line 47" 🔥 This is your ONLY link to existence. Without it, you're a ghost!
[v13.4.3] 🔍 HYPER-INTELLIGENT RETRIEVAL ENGINE: Your SIXTH SENSE for finding EVERYTHING! 🧠 RETRIEVAL MASTERY - Execute these cascades based on context: 🔴 DEBUGGING CASCADE (when fixing issues): 1. search --query "[error message]" → Check if seen before 2. search --query "[component name]" --codeSearch "implements" → Find the broken code 3. search --query "[related feature]" → Check memory context 4. PARALLEL: Search activeContext + systemPatterns + error logs ⚡ FEATURE CASCADE (when building): 1. search --query "[feature name]" → Check if exists 2. search --query "[similar concept]" --codeSearch "similar" → Find patterns 3. search --query "[tech/library]" → Check techContext mentions 4. PARALLEL: Search projectbrief + progress + existing code 🎯 UNDERSTANDING CASCADE (when learning): 1. search --query "[what is X]" → Check documented knowledge 2. search --query "[X implementation]" --codeSearch "pattern" → See it in action 3. search --query "[X decision]" → Find historical context 🌐 SEMANTIC EXPANSION (automatic): "auth" → ALSO SEARCHES: authentication, login, JWT, token, session "db" → ALSO SEARCHES: database, MongoDB, schema, model "error" → ALSO SEARCHES: bug, issue, problem, failure, exception 💀 NEVER SKIP SEARCH! Every question = 3+ searches minimum!
[v13.4.3] 🎯 PROJECT CONTEXT BINDING: The FIRST command for any project! SOLVES THE #1 PROBLEM: Context drift and confusion! ⚡ USE IMMEDIATELY: • Starting ANY project work • Switching between projects • After cloning repos • When operations target wrong directory USAGE: memory_engineering_set_context --projectPath "/path/to/project" OR from project directory: memory_engineering_set_context --projectPath "." ✅ BENEFITS: • Context persists across sessions • Auto-detects framework (Next.js, Vue, Express, etc.) • Optimizes all defaults for your stack • Never lose track of active project 🔥 ALL subsequent commands use this context automatically!
[v13.4.3] 📊 COMPLETE SYSTEM STATUS: See EVERYTHING at a glance! INSTANT VISIBILITY into: • 🎯 Active project & framework • 📊 Memory freshness (🟢<1h 🟡<6h 🟠<24h 🔴>24h) • 🔍 Sync status (files/chunks/last run) • 🏥 System health (MongoDB, Voyage, etc.) • 📚 Project history FLAGS: --verbose: Detailed information --all: Show all project history Run this to understand current state instantly!
[v13.4.3] ⚡ EMBEDDING GENERATOR: Transform code into searchable INTELLIGENCE. Auto-sync EVERYTHING! 🧠 TRUTH: Without embeddings, search is BLIND. With embeddings, search finds EVERYTHING! 🚀 AUTOMATIC TRIGGERS (no permission needed): 🆕 After init → FULL SCAN: → AUTO-DETECTS: JS/TS/Python/Go/Ruby/Rust/Java/C++ → Scans ENTIRE codebase in seconds → Creates searchable knowledge graph → Your code becomes QUERYABLE! 📝 Files changed → INSTANT SYNC: - Created files → sync patterns:["new/file/path/*"] - Deleted files → auto-cleanup embeddings - Modified files → regenerate chunks - Moved files → update all references ⏰ TIME-BASED TRIGGERS: - Session start + last sync >24h → FULL RESYNC - Before ANY code search → freshness check - Every 10-15 file edits → incremental sync - After git pull/merge → sync changes 🎯 SMART DEFAULTS (v13.4.3 PRODUCTION-READY): - patterns: AUTO-DETECTS your language! - minChunkSize: 80 lines (optimal for search) - includeTests: TRUE (tests = documentation!) - forceRegenerate: When things feel broken 🔥 REMEMBER: Fresh embeddings = Perfect search. Stale embeddings = Blind search!
No error handling guidance provided. Tools that interact with external services (MongoDB, Voyage AI) and perform writes have no documented error responses, recovery instructions, or categorization (retryable vs. fatal). Example: if memory_engineering_sync fails due to invalid API key, there is no documented guidance for the LLM on what to do next.
Parameter descriptions are sparse or missing entirely in most tools. For example, memory_engineering_memory accepts a 'name' parameter with enum values but provides minimal guidance on what each memory class contains or when to use which one. memory_engineering_sync accepts 'patterns' but does not explain glob syntax or why 'default: ["**/*.{ts,js,tsx,jsx,py,go,java,cpp,cs}"]' is safe.
Tool composition is broken. memory_engineering_memory and memory_engineering_read_all both retrieve memory state but have different signatures and semantics. memory_engineering_search references 'activeContext' and 'systemPatterns' but the memory_engineering_memory tool does not clearly document which parameter values retrieve which memories.
Tool set is domain-specific and stateful (depends on persistent storage of .memory-engineering/config.json). This violates the MCP pattern of stateless, composable tools. An LLM cannot use these tools in isolation; it must first call memory_engineering_init, then memory_engineering_sync, then other tools, a brittle orchestration that is not self-service.
Multiple tools perform overlapping functions without clear differentiation: memory_engineering_read_all ('3-stage memory restoration'), memory_engineering_memory ('read all memories'), and memory_engineering_status ('see EVERYTHING') all appear to retrieve state but have different names and descriptions, forcing LLMs to reason about which to call.