MCP server

Wire Mushi into Claude Desktop, Cursor, Codex CLI, or any Model Context Protocol client. The server exposes Mushi reports as tools, letting an agent triage, comment, and dispatch fixes from your IDE.

Install


pnpm add -g @mushi-mushi/mcp

Add to Claude Desktop (stdio transport — default)

~/Library/Application Support/Claude/claude_desktop_config.json


{
  "mcpServers": {
    "mushi-mushi": {
      "command": "mushi-mushi-mcp",
      "env": {
        "MUSHI_PROJECT_ID": "YOUR_PROJECT_ID",
        "MUSHI_API_KEY": "YOUR_SERVICE_API_KEY"
      }
    }
  }
}

Hosted Streamable HTTP transport (no subprocess)

The Mushi backend also exposes the same tool catalog over the Streamable HTTP transport from the MCP 2025-03-26 spec at /functions/v1/mcp. Use this when you’d rather not spawn a local subprocess — typical for OpenAI Agents SDK, ChatGPT Agent, hosted CrewAI, or any orchestrator that talks remote MCP:

.cursor/mcp.json


{
  "mcpServers": {
    "mushi-mushi-hosted": {
      "url": "https://api.mushimushi.dev/functions/v1/mcp",
      "headers": {
        "X-Mushi-Api-Key": "mushi_live_…",
        "X-Mushi-Project-Id": "proj_…"
      }
    }
  }
}

The endpoint accepts JSON-RPC 2.0 over POST (returns application/json or text/event-stream per content negotiation), opens an SSE stream on GET for server-pushed notifications, and accepts DELETE for session termination. Auth is the same dual-mode API-key / JWT used everywhere else on /v1/admin/*.

Available tools

Tool	Purpose
`list_reports`	Filter by status, severity, component
`get_report`	Full detail incl. screenshot + classification
`comment_on_report`	Post a triage note (audit-logged)
`transition_report`	`new → fixing → fixed` etc.
`dispatch_fix`	Hand off to the agentic fix orchestrator. Optionally accepts `inventoryActionNodeId` to pin the fix to a specific inventory `Action` (whitepaper §2.10 spec-traceability)
`get_fix_context`	One-shot brief for an external agent — root cause, smallest file set, repro, and the linked inventory `Action` with its `expected_outcome` contract when set
`query_knowledge_graph`	Ask “what other reports look like this?”

🔁

The MCP server speaks JSON-RPC 2.0 with tools/call semantics, and the agent surface is also discoverable via the A2A Agent Card at /.well-known/agent-card.