MCPOne-click Conductor in Claude Desktop / Cursor

Model Context Protocol (MCP) is an open protocol designed by Anthropic that lets AI assistants invoke external tools and read external resources. Think of it as "USB-C for AI" — one standard interface connecting any compatible model client to any tool service.

Nexevo's MCP server lives at https://api.nexevo.ai/v1/mcp(HTTP JSON-RPC 2.0, Bearer auth). Once enabled, inside Claude Desktop / Cursor you can:

• Tell Claude "push this design to Conductor and let GPT-5 review it" → invokes conductor_ask mode=chat
• "Save that auth decision to Recall, tag: auth" → invokes conductor_ask mode=save_memory
• "Find my prior capsule about webhook retries" → invokes conductor_ask mode=search_memory (semantic search)

The Nexevo MCP server exposes a single tool: conductor_ask. The mode parameter controls its behavior:

prompt:      string  // 你想让另一个 AI 处理的内容
template?:   string  // 可选模板:find_bug | counter_argument | fact_check | risk_assessment
judge_model?: string // 可选指定 model(如 "gpt-5" / "claude-opus-4-7");不传走 Conductor 自动选
options?:    object  // 可选 ConductorOptions(verify / agent / max_cost_usd 等)

title:   string         // 胶囊标题,如 "Clerk 选型决策"
content: string         // markdown 正文
kind?:   "chat" | "decision" | "manual"  // 默认 chat
tags?:   string[]       // 如 ["auth", "decision"]

query:  string         // 检索关键词(中英文皆可)
top_k?: number          // 默认 3,最多 20
tag?:   string          // 可选 tag 精确过滤

Tip: Claude Desktop / Cursor usually picks mode from natural language intent — say "have another AI review this" → mode=chat; "save this" → mode=save_memory. You don't need to specify mode explicitly.

Besides the conductor_ask tool, the MCP server exposes a resource URI scheme:

• workbench://session/{id} — Full JSON of a Quorum session (left + right messages + push events + verdict). MCP clients callresources/list to get your session list, thenresources/read for individual ones.

Typical use: tell Claude "look at my Clerk vs Auth0 Quorum from yesterday"; Claude calls resources/list → finds the matching session → resources/read pulls full dialog.

Nexevo MCP is an HTTP server; Claude Desktop primarily supports stdio MCP, so we use mcp-remote as a bridge (npm package, auto-fetched via npx).

Open ~/Library/Application Support/Claude/claude_desktop_config.json(Mac) or %APPDATA%\Claude\claude_desktop_config.json (Windows), and add:

{
  "mcpServers": {
    "nexevo": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://api.nexevo.ai/v1/mcp",
        "--header", "Authorization: Bearer YOUR_NEXEVO_API_KEY"
      ]
    }
  }
}

Get an API key: log in →Settings → API Keys→ create a new key (sk-...).

Save the config and fully quit Claude Desktop, then restart(not just close window — Cmd+Q / quit from taskbar). After restart, the tool 🔌 icon at bottom-right of Claude's input should shownexevo server listing conductor_ask.

Cursor MCP config lives in ~/.cursor/mcp.json or per-project.cursor/mcp.json; JSON structure same as Claude Desktop:

{
  "mcpServers": {
    "nexevo": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://api.nexevo.ai/v1/mcp",
        "--header", "Authorization: Bearer YOUR_NEXEVO_API_KEY"
      ]
    }
  }
}

Settings → Tools & Integrations → MCP servers; verify nexevo showsconductor_ask tool. Cursor Composer / Chat will invoke it naturally.

MCP protocol is JSON-RPC 2.0 over HTTP POST. Any HTTP client can hit it:

# 1. initialize 拿 server info
curl https://api.nexevo.ai/v1/mcp \
  -H "Authorization: Bearer $NEXEVO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "jsonrpc": "2.0", "id": 1, "method": "initialize" }'

# 2. 列 tools(应只见 conductor_ask)
curl https://api.nexevo.ai/v1/mcp \
  -H "Authorization: Bearer $NEXEVO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "jsonrpc": "2.0", "id": 2, "method": "tools/list" }'

# 3. 调 conductor_ask mode=save_memory(存胶囊)
curl https://api.nexevo.ai/v1/mcp \
  -H "Authorization: Bearer $NEXEVO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 3,
    "method": "tools/call",
    "params": {
      "name": "conductor_ask",
      "arguments": {
        "mode":    "save_memory",
        "title":   "Clerk vs Auth0 决策",
        "content": "选 Clerk,理由:5 行集成 + 免费层 10k MAU + UI 现代化。",
        "tags":    ["auth", "decision"]
      }
    }
  }'

# 4. 调 conductor_ask mode=search_memory(语义召回)
curl https://api.nexevo.ai/v1/mcp \
  -H "Authorization: Bearer $NEXEVO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 4,
    "method": "tools/call",
    "params": {
      "name": "conductor_ask",
      "arguments": {
        "mode":  "search_memory",
        "query": "用户认证方案",
        "top_k": 3
      }
    }
  }'

# 5. 调 conductor_ask mode=chat(让另一个 AI 帮审)
curl https://api.nexevo.ai/v1/mcp \
  -H "Authorization: Bearer $NEXEVO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 5,
    "method": "tools/call",
    "params": {
      "name": "conductor_ask",
      "arguments": {
        "mode":     "chat",
        "prompt":   "审一下这段 JWT 验证代码: ...",
        "template": "find_bug",
        "judge_model": "gpt-5"
      }
    }
  }'

• Conductor doc — conductor_ask mode=chat invokes the Conductor pipeline internally
• Recall doc — mode=save_memory / search_memory operate on Recall capsules
• Settings → API Keys — create / revoke / see last used
• MCP protocol official docs (modelcontextprotocol.io)
• mcp-remote (stdio ↔ HTTP bridge)