@mushi-mushi/mcp

MCP server. See Quickstart → MCP for setup.

Speaks JSON-RPC 2.0 with tools/call semantics and supports the SEP-1686 Tasks envelope so long-running fixes can stream progress back into your IDE.

Two transports

• stdio (local) — npx -y @mushi-mushi/mcp@latest. The default for Cursor / Claude Desktop / Claude Code / Continue / Cline / Zed / Windsurf. Your client launches the server as a subprocess and pipes JSON-RPC over stdio.
• Streamable HTTP (hosted, 2026-05-09 release) — single endpoint at /functions/v1/mcp per the MCP 2025-03-26 spec. POST returns application/json or text/event-stream per content negotiation, GET opens an SSE stream for server-pushed notifications, DELETE terminates the session. Use this when you want to skip the local subprocess (typical for OpenAI Agents SDK, ChatGPT Agent, hosted CrewAI, or any orchestrator that talks remote MCP).

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

Auth on both transports is the same dual-mode (API key with mcp:read|write OR JWT) used everywhere on /v1/admin/*.

Spec traceability for dispatch_fix

The dispatch_fix tool now accepts inventoryActionNodeId so callers that already know the inventory Action they want repaired can pin the fix directly. The complementary get_fix_context tool surfaces the linked Action (with expected_outcome) so an external orchestrator can read the contract before drafting a fix.

// dispatch_fix tool inputSchema (excerpt)
{
  "type": "object",
  "required": ["reportId"],
  "properties": {
    "reportId": { "type": "string" },
    "projectId": { "type": "string" },
    "inventoryActionNodeId": { "type": "string" }
  }
}

See @mushi-mushi/mcp on npm for the full tool catalog and Connecting your orchestrator for the non-MCP surfaces (REST, A2A, AG-UI SSE).

Discoverable via the A2A Agent Card.