v0.2.0 — Fallback Routing & Usage Statistics

Phase 3: Orchestration

New Features

• Fallback Routing: When a tool call fails on one server, automatically retries on other servers that provide the same tool. Candidates are sorted by success rate (when preferHealthyServers=true) then by priority.
• Usage Statistics: Every tool call is recorded with success/failure status, response time, and fallback usage. View via the new mcp_stats tool.
• mcp_stats Tool: View usage statistics with detail parameter:
  • summary (default): overall stats + recent calls
  • server: per-server breakdown (calls, success rate, avg response time, fallbacks)
  • tool: per-tool breakdown (calls, success rate, servers used)
  • all: everything
• Priority-weighted Sorting: Servers with higher priority are preferred. When preferHealthyServers=true, success rate takes precedence over priority.
• preferredServer Override: Users can specify which server to try first via the serverName parameter in mcp_call_tool.

New Config Fields

Field	Default	Description
enableFallback	true	Enable fallback routing
maxFallbackAttempts	3	Max servers to try
preferHealthyServers	true	Sort by success rate (false = priority only)
enableStats	true	Register mcp_stats tool

Test Results

Spike:         34 passed, 0 failed
Integration:  236 passed, 0 failed
Total:        270 passed, 0 failed

Install

dsh plugin --profile web add github:Daseanle/dsh-mcp-orchestrator

v0.1.0 — MVP

dsh-mcp-orchestrator v0.1.0

MCP orchestration layer for DeepSeek Harness.

Features

• Multi-server MCP stdio connections
• Auto-restart with exponential backoff
• Health monitoring (ping-based)
• Tool collision detection and 3 namespace modes (collision/server/dot)
• 3 DSH management tools: mcp_list_servers, mcp_list_tools, mcp_call_tool
• Zod config schema with cordis.patch.yml support

Test Results

• Spike: 34 passed, 0 failed
• Integration: 124 passed, 0 failed
• Total: 158 passed, 0 failed

Install

dsh plugin --profile web add github:Daseanle/dsh-mcp-orchestrator