OverMind-MCP exposes 23 tools via fastmcp/HTTP. Descriptions exist but many are generic or vague. Schemas are present but lack clarity on required vs optional fields, and some parameters lack proper type constraints. The tool set mixes domain concerns (agent orchestration, A2A communication, privacy compliance) without clear separation. French descriptions create accessibility issues for international LLM usage. Error handling guidance is absent, no indication of retryability, recovery paths, or what to do when a tool fails. Naming is inconsistent (run_agent vs run_agents_parallel uses different conventions). Critical security tools (loi25_*) exist but lack descriptions explaining when/why to use them vs the core agent tools.

• Translate all tool and parameter descriptions to English. Use clear, concise descriptions (50-200 chars) following LLM-optimized patterns (WHAT it does, WHEN to use it, WHAT it returns).
• Add complete JSON Schema definitions for all parameters. Mark required fields explicitly. For object parameters (e.g., 'updates', 'metadata', 'riskAssessment'), define nested schemas with allowed fields and types.
• Split a2a_hub into five separate tools: a2a_broadcast, a2a_pipeline, a2a_fanout, a2a_delegate, a2a_query. Each tool should have a clear, focused description and enum-constrained mode parameter removed.
• Add error handling and recovery guidance to every tool description. Example: 'If the agent is busy, wait 2 seconds and retry. If offline, check runner status with agent_control(agentName, "status").'
• Implement confirmation flow for destructive tools (delete_agent, loi25_erasure). Add a 'confirm' boolean parameter and return a confirmation request (MRTR pattern: result type 'input_required') before executing irreversible operations.
• Document output schemas for all tools. Include example responses showing field names, types, and which IDs can be passed to downstream tools. Use the response-shaper pattern to strip unnecessary metadata.
• Consolidate run_agent and run_agents_parallel into a single run_agent tool with an 'agents' array parameter (single item or batch). Remove the plural variant to reduce LLM confusion.
• Declare 'runner', 'mode', and 'consentType' parameters as enums rather than free-form strings. Include enum values in schema and parameter description.

Inferred effective spec: <=2025-11-25.

• Relies on Sampling (deprecated) - integrate directly with the LLM provider API