Loading ToolBench
Cross-platform AI cognitive capture via MCP — stores, analyzes, and retrieves cognitive events from multiple AI platforms with coherence detection, knowledge graphs, and cryptographic proof-of-cognition
UCW has 20 tools with complete schemas and descriptions, but quality is uneven. Naming is verb-forward (store_, search_, get_, coherence_, verify_, etc.), which is good. Descriptions average ~120 chars and are present for all tools. However, many descriptions lack actionable context: they state WHAT the tool does but not WHEN to use it or what prerequisites exist. Parameters have types and descriptions, but several lack constraints (enums, ranges). Output schemas are not documented, LLMs cannot predict return structure. Error handling is minimal; tools return generic error messages without recovery guidance. Security concerns: no mention of permission gates or audit trails for destructive operations (store_learning, coherence_scan, link_threads, generate_receipt). Composition is reasonable, tools are single-purpose, but some are tightly coupled to internal state (coherence_scan auto-saves to coherence_moments; generate_receipt builds Merkle trees). Overall: functional but not production-grade.
Query intelligence alerts with optional filters by type, severity, and acknowledgment status. Returns alerts in reverse chronological order.
List high-coherence events with emergence indicators. Shows breakthrough potential, meta-cognitive events, and concept clusters.
Scan recent events for coherence patterns and emergence signals. Breakthrough clusters are auto-saved to coherence_moments table. Returns a summary of detected patterns.
Semantic similarity search across captured events using embeddings. Finds events similar to a natural language query.
Get UCW Coherence Engine status: total events, session count, topic distribution, and gut signal breakdown.
Find coherence signatures appearing on 2+ platforms. Detects when the same cognitive pattern emerges across different AI tools.
Output schemas not documented. LLMs cannot predict return structure (fields, types, pagination). Tools return text_content() responses without structured field definitions.
Descriptions lack actionable context. Most state WHAT the tool does but not WHEN to use it, prerequisites, or how it differs from similar tools. E.g., 'Get UCW Coherence Engine status' does not explain when to call it vs coherence_moments or coherence_search.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | <=2025-11-25 | v2 |
Generate a cryptographic proof-of-cognition receipt for a session. Builds a Merkle tree over event chain hashes and stores the receipt.
Get relevant context for a query from stored learnings and recent cognitive events
Analyze entity clusters and relationship patterns in the knowledge graph
Query the UCW knowledge graph: search entities, get relationships, or view graph statistics
Trigger thread linking on recent captured events. Scans the last N events and groups them into conversation threads.
Show proof-of-cognition status: chain length, receipts generated, integrity check, and Merkle root for the current session.
Search stored agent learnings by query, project, or topic
Track coherence progression for a specific topic over time. Shows how understanding deepens day by day.
Store a learning or insight for future retrieval. Agents can write back knowledge discovered during sessions.
Combined temporal analysis: knowledge decay detection, activity heatmap, and overall temporal stats
Analyze cross-platform conversation threads. Actions: 'list' (all threads), 'cross_platform' (multi-platform), 'stats' (summary).
Track how topics evolve over time. Shows daily topic frequency grouped by day.
Verify the hash chain integrity for a session. Checks that every cognitive event links correctly to its predecessor, detecting any tampering or data corruption.
Verify a previously generated proof-of-cognition receipt. Rebuilds the Merkle tree from current events and compares roots.
No permission gates or audit trails for destructive operations. store_learning, coherence_scan, link_threads, and generate_receipt modify state but lack permission checks or logging guidance. No mention of who called what, when, or with what parameters.
Error handling is generic and non-actionable. Tools return 'Error in {name}: {exc}' without recovery guidance. LLMs cannot determine if errors are retryable, user-fixable, or fatal.
Parameters lack constraints. 'limit' and 'days' parameters have no min/max bounds. 'action' enums exist (e.g., thread_analysis, knowledge_graph) but other string parameters (query, topic, entity_type) are free-form, inviting hallucinated values.