Loading ToolBench

Reading the current corpus…

MemGuard MCP server · D · ToolBench

MemGuard

Exposes MemGuard validation tools that AI agents can call natively via the Model Context Protocol (MCP). Tools allow agents to check memory trustworthiness before acting on stored facts.

2026-07-28+STDIO4 tools

59/100

Definition Quality77

Protocol Readiness50

Supportability34

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

MemGuard exposes 4 tools with complete schemas and descriptions. Naming follows verb_noun convention (validate_, get_, report_, get_). Descriptions are substantive (100-200 chars) and explain WHEN to use each tool. However, output schemas are undocumented, LLMs cannot predict response structure. Error handling is absent from visible code. Parameters lack validation constraints (e.g., min_trust_score 0.0-1.0 is stated but not enforced). No tool annotations (readOnlyHint, destructiveHint). The server is single-tenant hardcoded, limiting production readiness.

Tools (4)

get_memory_healthread onlyauthsource verified73/100

Get overall health metrics for the agent's memory store. Use this to assess how reliable the current memory state is.

get_trusted_memoriesread onlyauthsource verified75/100

Retrieve only memories above a trust score threshold. Use instead of raw memory retrieval when accuracy is critical.

report_stale_memorywriteauthsource verified80/100

Report a memory that the agent suspects is stale based on new information encountered during a task. This triggers priority validation.

validate_memoryread onlyauthsource verified78/100

Check if a specific memory is still accurate before acting on it. Returns trust score and validation status. Call this before making decisions based on stored facts that might be outdated.

Issues (5)

HIGHtool ↗

Output schemas undocumented. LLMs cannot predict response structure (fields, types, pagination). Forces agents to guess what data is returned.

validate_memoryget_memory_healthreport_stale_memoryget_trusted_memories

HIGHrecovery-guide ↗

No error handling guidance in visible code. Agents cannot distinguish retryable errors from fatal ones. No recovery hints (e.g., 'memory_id not found, try search_memories first').

validate_memoryget_memory_healthreport_stale_memoryget_trusted_memories

MEDIUMtool-annotations ↗

No tool annotations. Missing readOnlyHint (get_* tools), destructiveHint (report_stale_memory writes), idempotentHint. Agents cannot reason about side effects.

validate_memoryget_memory_healthreport_stale_memoryget_trusted_memories

HIGHpermission-gate ↗

Recommendations

Document output schemas for all 4 tools. Example: validate_memory returns {trust_score: number, validation_status: string, last_validated: ISO8601, strategy_used: string}. Include pagination for list results (total_count, next_cursor).
Add tool annotations: readOnlyHint=true for validate_memory, get_memory_health, get_trusted_memories; destructiveHint=true for report_stale_memory (triggers validation job).
Implement error handling with recovery guidance. E.g., if memory_id not found, return 'Memory not found. Try get_trusted_memories(query=...) to search by content.' Categorize errors as retryable (validation service timeout) vs user-fixable (invalid UUID format).
Replace hardcoded DEFAULT_TENANT_ID with MCP auth context resolution. Extract tenant_id from request metadata or MCP session. Document required permissions (e.g., 'read:memory', 'write:validation').
Add input validation: min_trust_score must be 0.0 - 1.0; limit must be 1 - 100; strategy must be one of [quick, source_linked, semantic]. Return actionable errors: 'Invalid min_trust_score: 1.5, must be 0.0 to 1.0.'

Score history

Overall score trend

First recorded score · v2 rubric

59/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	D	59	2026-07-28+	v2