/ 目錄 / 演練場 / GPT Researcher
● 官方 assafelovic 🔑 需要你的金鑰

GPT Researcher

作者 assafelovic · assafelovic/gpt-researcher

Autonomous research agent: give it a question, it plans searches, reads sources, and writes a cited report — callable from Claude via MCP.

GPT Researcher decomposes a research question into sub-queries, fetches web and document sources, and synthesizes a cited report. Its MCP server lets Claude offload deep research while staying in the chat.

為什麼要用

核心特性

即時演示

實際使用效果

就緒

安裝

選擇你的客戶端

~/Library/Application Support/Claude/claude_desktop_config.json  · Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "gpt-researcher-mcp": {
      "command": "uvx",
      "args": [
        "gpt-researcher-mcp"
      ]
    }
  }
}

開啟 Claude Desktop → Settings → Developer → Edit Config。儲存後重啟應用。

~/.cursor/mcp.json · .cursor/mcp.json
{
  "mcpServers": {
    "gpt-researcher-mcp": {
      "command": "uvx",
      "args": [
        "gpt-researcher-mcp"
      ]
    }
  }
}

Cursor 使用與 Claude Desktop 相同的 mcpServers 格式。專案級設定優先於全域。

VS Code → Cline → MCP Servers → Edit
{
  "mcpServers": {
    "gpt-researcher-mcp": {
      "command": "uvx",
      "args": [
        "gpt-researcher-mcp"
      ]
    }
  }
}

點擊 Cline 側欄中的 MCP Servers 圖示,然後選 "Edit Configuration"。

~/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "gpt-researcher-mcp": {
      "command": "uvx",
      "args": [
        "gpt-researcher-mcp"
      ]
    }
  }
}

格式與 Claude Desktop 相同。重啟 Windsurf 生效。

~/.continue/config.json
{
  "mcpServers": [
    {
      "name": "gpt-researcher-mcp",
      "command": "uvx",
      "args": [
        "gpt-researcher-mcp"
      ]
    }
  ]
}

Continue 使用伺服器物件陣列,而非映射。

~/.config/zed/settings.json
{
  "context_servers": {
    "gpt-researcher-mcp": {
      "command": {
        "path": "uvx",
        "args": [
          "gpt-researcher-mcp"
        ]
      }
    }
  }
}

加入 context_servers。Zed 儲存後熱重載。

claude mcp add gpt-researcher-mcp -- uvx gpt-researcher-mcp

一行命令搞定。用 claude mcp list 驗證,claude mcp remove 移除。

使用場景

實戰用法: GPT Researcher

Run a deep research task from inside Claude

👤 Developers ⏱ ~15 min intermediate

何時使用: Question needs 20+ sources and a written summary, not 3 web hits.

步驟
  1. Bridge
    Add GPT Researcher MCP.✓ 已複製
    → Tool visible
  2. Commission
    Ask Claude to commission the research.✓ 已複製
    → Researcher runs
  3. Receive
    Cited report comes back.✓ 已複製
    → Ready to use

結果: Cited multi-source report in one prompt.

組合

與其他 MCP 搭配,撬動十倍槓桿

gpt-researcher-mcp + exa-mcp-server

Use Exa for fast lookups; GPT Researcher for deep multi-source work

Combine gpt-researcher-mcp with exa-mcp-server: Use Exa for fast lookups; GPT Researcher for deep multi-source work✓ 已複製

工具

此 MCP 暴露的能力

工具輸入參數何時呼叫成本
research.run (see docs) Run a multi-step research task 1 call
research.status (see docs) Check on a running task 1 call

成本與限制

運行它的成本

API 配額
Per LLM + search provider
每次呼叫 Token 數
Heavy — long contexts
費用
Free OSS; LLM + search costs
提示
Set max-iterations to cap runaway researchers

安全

權限、密鑰、影響範圍

最小權限: search:read
憑證儲存: API keys in env
資料出站: Configured providers
切勿授予: Send proprietary data unless self-hosting

故障排查

常見錯誤與修復

Researcher loops on same sources

Lower iteration cap; tighten the question

替代方案

GPT Researcher 對比其他方案

替代方案何時用它替代權衡
Manual web + search MCPOne-off questionsNo synthesis layer

更多

資源

📖 閱讀 GitHub 上的官方 README

🐙 查看未解決的 issue

🔍 瀏覽全部 400+ MCP 伺服器和 Skills