Loading ToolBench
Local MCP server for Minty network memory. Implements MCP JSON-RPC over stdio for Hermes native MCP client. Read-only, no LLM calls, no outreach, source-backed outputs only.
Minty MCP server has well-structured tool definitions with consistent naming (verb_noun pattern: search_network, person_context, workflow_brief, etc.) and comprehensive descriptions (150-250 chars each). All 7 tools have input schemas with typed parameters and descriptions. However, output schemas are not documented in the visible code, parameter constraints lack enums where applicable (e.g., source/sources accept free-form strings instead of enums), and error handling guidance is absent. All tools are read-only (good for safety), but descriptions could be more explicit about what data is returned and when to use each tool vs. alternatives. No tool annotations (readOnlyHint) present despite all being read-only operations.
Recommend privacy-safe next actions for active relationship goals. Prioritizes active pipeline follow-ups before new asks and warm-intro paths. Read-only — no messages sent, no outreach triggered, no contacts mutated.
Find privacy-safe warm intro paths to a target person or goal using local group co-membership evidence. Returns redacted intermediaries, shared-context counts, citations, and safety metadata. Read-only — no messages sent, no outreach triggered, no contacts mutated.
Prepare a privacy-safe brief for an upcoming calendar meeting. Returns opaque refs, attendee relationship context, citations, freshness, and safety metadata. Read-only — no calendar changes, messages, or outreach.
Get context about a specific person in your network. Returns matching contacts with relationship evidence, topics, and warmth. No emails, phones, or raw contact details included.
Search your private network for people matching a natural-language query. Returns ranked results with relationship evidence, warmth, and suggested actions. Read-only — no messages sent, no contacts mutated. Contact details (emails/phones) omitted by default.
Output schemas not documented. Tools return complex objects (ranked results, relationship evidence, citations, metadata) but LLMs cannot see the response structure. This forces agents to guess at available fields and risks failed downstream chaining.
source/sources parameters accept free-form strings instead of enums. Valid values (telegram, whatsapp, linkedin, slack, discord, email, sms, googlecontacts) are documented in descriptions but not enforced as enums. LLMs may hallucinate invalid source names.
No error handling guidance. Tools lack descriptions of failure modes, recovery steps, or actionable error messages. If a query returns no results or a source is unavailable, agents have no guidance on what to try next.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
Check which Minty sources are fresh, evidence-bearing, and safe for source-specific agent queries. Read-only and redacted. No contact details, raw rows, or token paths exposed.
Generate a concise brief for a Hermes workflow goal. Returns top relevant people, why each matters, suggested safe next steps, and data freshness/safety metadata. Read-only, source-backed.
No tool annotations. All 7 tools are read-only (safe to retry, no side effects) but lack readOnlyHint annotations. This forces LLMs to infer safety from descriptions rather than machine-readable metadata.
Ambiguous parameter relationships. Tools like search_network accept both 'source' (singular) and 'sources' (array), the interaction is undocumented. Do they combine? Does one override the other? LLMs will guess.