Loading ToolBench
MCP Server for Agent Context Graph — exposes agent_context, agent_status, and agent_list tools via JSON-RPC over stdio. Resolves natural-language task queries to relevant Agent Graph nodes, returns matched agents, skills, files to read, and token estimates.
Three tools with complete schemas and reasonable descriptions. Tool naming follows verb_noun convention (gitnexus_agent_*). However, descriptions lack actionable context for LLM selection, they state WHAT but not WHEN or WHY to use each tool. Parameter descriptions are present but generic. No output schema documentation. Error handling and recovery guidance absent. Schemas are well-formed JSON Schema with types and enums, but lack constraints (min/max for depth/max_tokens). No tool annotations (readOnlyHint, idempotentHint). Composition is sound, three distinct, single-purpose tools.
Resolve a natural-language task query to relevant Agent Graph nodes. Returns matched agents, skills, files to read, and token estimates. Use this BEFORE reading context files to minimize token usage. Average savings: 93% of context window.
List all agents and skills in the Agent Graph. Returns agent names, roles, and their associated skills.
Show Agent Graph statistics: node counts by type, edge counts, and index freshness. Use this to check if the agent graph is healthy.
No output schema documentation. LLMs cannot plan downstream calls or extract required fields. gitnexus_agent_context returns ContextResult (version, query, matched_agents, matched_skills, context_chain, files_to_read, estimated_tokens, savings_vs_full, metadata) but this is not declared in the tool definition.
Descriptions lack WHEN/WHY context. 'Resolve a natural-language task query...' states WHAT but not when to call this vs. gitnexus_agent_list. LLMs need explicit guidance: 'Use this BEFORE reading context files to minimize token usage' is present but buried. Add: 'Call this when you have a task description and need to find relevant agents/skills. Use gitnexus_agent_list to browse all available agents first.'
Numeric parameters lack constraints. 'depth' (default 2, max 5) and 'max_tokens' (default 5000) have no min/max in schema. Add minValue/maxValue: depth {minimum: 1, maximum: 5}, max_tokens {minimum: 100, maximum: 50000}.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | <=2025-11-25 | v2 |
No error handling guidance. If gitnexus_agent_context fails (e.g., database not found, query matches no agents), the tool provides no recovery hint. Add: 'If no agents match, try a broader query or call gitnexus_agent_list to see available agents.'
No tool annotations. All three tools are read-only and idempotent but lack readOnlyHint/idempotentHint in schema. This prevents agents from reasoning about safety and retry logic.