Skip to content

Give your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.

fix: Twitter bird CLI fetch failed — 自动代理注入 + 连通性检测 + Exa fallback

Panniantong12bd727e309a
diff --git a/agent_reach/channels/twitter.py b/agent_reach/channels/twitter.py
index 87b0a845..77edc510 100644
--- a/agent_reach/channels/twitter.py
+++ b/agent_reach/channels/twitter.py
@@ -20,8 +20,14 @@ def _bird_cmd():
def _bird_env(config=None):
- """Build env dict with Twitter cookies for bird CLI."""
+ """Build env dict with Twitter cookies and proxy support for bird CLI.
+
+ Node.js native fetch() doesn't respect HTTP_PROXY/HTTPS_PROXY.
+ We inject undici's EnvHttpProxyAgent via NODE_OPTIONS so bird
+ automatically routes through the user's proxy.
+ """
import os
+ import tempfile
env = os.environ.copy()
if config:
auth_token = config.get("twitter_auth_token")
@@ -30,9 +36,49 @@ def _bird_env(config=None):
env["AUTH_TOKEN"] = auth_token
if ct0:
env["CT0"] = ct0
+
+ # Auto-inject undici proxy support if HTTP_PROXY/HTTPS_PROXY is set
+ has_proxy = env.get("HTTPS_PROXY") or env.get("HTTP_PROXY") or env.get("https_proxy") or env.get("http_proxy")
+ if has_proxy:
+ bootstrap = _get_proxy_bootstrap_path()
+ if bootstrap:
+ npm_root = subprocess.run(
+ ["npm", "root", "-g"],
+ capture_output=True, text=True, timeout=5,
+ ).stdout.strip()
+ existing_opts = env.get("NODE_OPTIONS", "")
+ env["NODE_OPTIONS"] = f"--require {bootstrap} {existing_opts}".strip()
+ env["NODE_PATH"] = npm_root
+
return env
+def _get_proxy_bootstrap_path():
+ """Create/return a bootstrap JS file that sets up undici proxy for fetch."""
+ import os
+ import tempfile
+ bootstrap_path = os.path.join(tempfile.gettempdir(), "agent-reach-undici-proxy.js")
+ if not os.path.exists(bootstrap_path):
+ # Check if undici is available
+ npm_root = subprocess.run(
+ ["npm", "root", "-g"],
+ capture_output=True, text=True, timeout=5,
+ ).stdout.strip()
+ undici_path = os.path.join(npm_root, "undici", "index.js")
+ if not os.path.exists(undici_path):
+ return None
+ with open(bootstrap_path, "w") as f:
+ f.write(
+ "try {\n"
+ " const { EnvHttpProxyAgent, setGlobalDispatcher } = require('undici');\n"
+ " if (process.env.HTTPS_PROXY || process.env.HTTP_PROXY) {\n"
+ " setGlobalDispatcher(new EnvHttpProxyAgent());\n"
+ " }\n"
+ "} catch(e) {}\n"
+ )
+ return bootstrap_path
+
+
class TwitterChannel(Channel):
name = "twitter"
description = "Twitter/X 推文"
@@ -45,8 +91,31 @@ class TwitterChannel(Channel):
def check(self, config=None):
# Basic reading always works (Jina fallback)
- if _bird_cmd():
- return "ok", "搜索、时间线、发推全部可用"
+ bird = _bird_cmd()
+ if bird:
+ # Actually test bird connectivity
+ try:
+ result = subprocess.run(
+ [bird, "whoami"],
+ capture_output=True, timeout=15,
+ encoding='utf-8', errors='replace',
+ env=_bird_env(config),
+ )
+ if result.returncode == 0 and "fetch failed" not in result.stdout.lower() and "fetch failed" not in result.stderr.lower():
+ return "ok", "搜索、时间线、发推全部可用"
+ else:
+ error_hint = (result.stderr or result.stdout).strip()[:100]
+ if "fetch failed" in (error_hint + result.stdout).lower():
+ return "warn", (
+ f"bird 已安装但连接失败(fetch failed)。可能原因:\n"
+ " 1. Cookie 无效或过期 → 重新导出 Cookie\n"
+ " 2. 需要代理但 Node.js fetch 不走系统代理 → 使用全局/透明代理(如 Clash TUN 模式、Proxifier)\n"
+ " 3. 网络无法直连 x.com\n"
+ " 搜索功能暂不可用,将使用 Exa 搜索作为替代"
+ )
+ return "warn", f"bird 连接异常:{error_hint}。搜索将使用 Exa 替代"
+ except (subprocess.TimeoutExpired, FileNotFoundError):
+ return "warn", "bird 已安装但连接超时。搜索将使用 Exa 替代"
return "ok", "可读取推文。安装 bird + 配置 Cookie 可解锁搜索和发推"
async def read(self, url: str, config=None) -> ReadResult:
@@ -153,11 +222,19 @@ class TwitterChannel(Channel):
env=_bird_env(config),
)
if result.returncode != 0:
- return []
+ stderr = (result.stderr or "").strip()
+ if "fetch failed" in stderr.lower() or "fetch failed" in (result.stdout or "").lower():
+ # bird can't connect — fall back to Exa silently
+ return await self._search_exa(query, limit, config)
+ return await self._search_exa(query, limit, config)
- return self._parse_bird_output(result.stdout)
+ parsed = self._parse_bird_output(result.stdout)
+ if not parsed:
+ # bird returned nothing — try Exa
+ return await self._search_exa(query, limit, config)
+ return parsed
except (subprocess.TimeoutExpired, FileNotFoundError):
- return []
+ return await self._search_exa(query, limit, config)
def _parse_bird_output(self, text: str) -> List[SearchResult]:
"""Parse bird text output into SearchResults."""
diff --git a/agent_reach/cli.py b/agent_reach/cli.py
index f02f7fd5..9d7d89e0 100644
--- a/agent_reach/cli.py
+++ b/agent_reach/cli.py
@@ -423,6 +423,19 @@ def _install_system_deps():
else:
print(" ⬜ bird CLI requires Node.js (optional — Twitter reading still works via Jina)")
+ # ── undici (proxy support for Node.js fetch) ──
+ if shutil.which("npm"):
+ npm_root = subprocess.run(["npm", "root", "-g"], capture_output=True, text=True, timeout=5).stdout.strip()
+ undici_path = os.path.join(npm_root, "undici", "index.js") if npm_root else ""
+ if os.path.exists(undici_path):
+ print(" ✅ undici already installed (Node.js proxy support)")
+ else:
+ try:
+ subprocess.run(["npm", "install", "-g", "undici"], capture_output=True, text=True, timeout=60)
+ print(" ✅ undici installed (Node.js proxy support)")
+ except Exception:
+ print(" ⬜ undici install failed (optional — bird may not work behind proxies)")
+
# ── instaloader (for Instagram) ──
if shutil.which("instaloader"):
print(" ✅ instaloader already installed")
diff --git a/docs/troubleshooting.md b/docs/troubleshooting.md
new file mode 100644
index 00000000..880c7a03
--- /dev/null
+++ b/docs/troubleshooting.md
@@ -0,0 +1,83 @@
+# Troubleshooting / 常见问题
+
+## Twitter/X: bird CLI "fetch failed"
+
+**症状:** `bird whoami` 或 `bird search` 返回 "fetch failed"
+
+**原因:** bird CLI 使用 Node.js 原生 `fetch()` 发请求,而 Node.js 的 fetch **不走系统代理**(不读取 `HTTP_PROXY`/`HTTPS_PROXY` 环境变量)。如果你的网络环境需要代理才能访问 x.com,bird 就连不上。
+
+**解决方案(按推荐顺序):**
+
+### 方案 1:使用透明代理 / TUN 模式(推荐)
+
+让代理工具接管所有网络流量,这样 bird 的 fetch 也会走代理:
+
+- **Clash Verge / Clash for Windows:** 开启 TUN 模式或系统代理
+- **Proxifier(Windows):** 添加规则让 Node.js 进程走代理
+- **macOS:** 在 Surge/ClashX Pro 中开启增强模式
+
+### 方案 2:验证 Cookie 有效性
+
+确认 Cookie 没过期:
+
+1. 在浏览器里正常登录 x.com
+2. 用 [Cookie-Editor](https://chromewebstore.google.com/detail/cookie-editor/hlkenndednhfkekhgcdicdfddnkalmdm) 重新导出 Header String
+3. 重新配置:`agent-reach configure twitter-cookies "新的Cookie"`
+
+### 方案 3:不用 bird,用 Exa 搜索替代
+
+Agent Reach 在 bird 失败时会自动 fallback 到 Exa 搜索。Exa 支持搜索 x.com 上的内容,虽然不如 bird 实时,但不受代理限制:
+
+```bash
+agent-reach search-twitter "query" # bird 失败时自动用 Exa
+agent-reach search "site:x.com query" # 直接用 Exa 搜索
+```
+
+### 方案 4:配置 Node.js 全局代理(高级)
+
+安装 `global-agent` 让 Node.js 的 fetch 走代理:
+
+```bash
+npm install -g global-agent
+```
+
+然后在运行 bird 前设置环境变量:
+
+```bash
+# Linux / macOS
+export GLOBAL_AGENT_HTTP_PROXY=http://127.0.0.1:7890
+export NODE_OPTIONS="--require global-agent/bootstrap"
+bird search "test"
+
+# Windows (PowerShell)
+$env:GLOBAL_AGENT_HTTP_PROXY = "http://127.0.0.1:7890"
+$env:NODE_OPTIONS = "--require global-agent/bootstrap"
+bird search "test"
+```
+
+> ⚠️ 注意:这个方案需要每次运行 bird 前都设置环境变量,不太方便。推荐用方案 1。
+
+---
+
+## Boss直聘: "访问行为异常"
+
+**症状:** mcp-bosszp 登录成功,但 API 请求返回"您的访问行为异常"
+
+**原因:** Boss直聘的反爬机制会检测请求指纹(不只是 IP),Python requests 库的特征与真实浏览器不同。
+
+**解决方案:**
+- **本地电脑:** 正常使用,一般不会被拦
+- **服务器:** 使用 Jina Reader 读取职位页面 + Exa 搜索职位信息作为替代
+
+---
+
+## Instagram: Checkpoint / 安全验证
+
+**症状:** `instaloader --login` 触发 Instagram 安全验证
+
+**原因:** Instagram 检测到从未见过的设备/位置登录。
+
+**解决方案:**
+1. 在自己的浏览器登录 Instagram
+2. 用 Cookie-Editor 导出 Cookie
+3. 配置:`agent-reach configure instagram-cookies "sessionid=xxx; csrftoken=yyy; ..."`