Building in public. Working on MCPSDK — tools & infrastructure for AI agents to actually ship.

Paris
Joined May 2025
Pinned Tweet
🚀 MCPSDK beta is live. We’re officially opening the waitlist today. Run 5000+ MCP Servers instantly — with one line of code. If integrating MCPs still means config files, Docker, permission headaches — this is for you. 👉 Join the waitlist: mcpsdk.dev
1
1
8
6
511
Back from two weeks off over the Mid-Autumn Festival and National Day holidays. Working from home today. Our SaaS team's agent platform runs 24/7 and has made a small profit for six months. Then OpenAI came out with dots. Small teams do feel the pressure.
1
35
Busabase is open source: use it locally, self-host it, or use Cloud. Product Hunt launch: tomorrow, Oct 9, 2026, at 00:00 PDT / 03:00 EDT / 07:00 UTC / 15:00 UTC+8. Same moment, all on Oct 9. I'm one of the developers on the team and would like to hear what you think.
1
19
Agents can do more and more. What should the business systems supporting that work look like? That's what we're trying to answer with Busabase. Come see our take on the shovel at busabase.com, and tell us what you think.
9
今天发生了一件让所有 AI 开发者后背发凉的事。 litellm,那个统一调用各家大模型 API 的 Python 库,GitHub 4 万星,月下载 9500 万次——被投毒了。 一行 pip install,你的 SSH 密钥、AWS/GCP/Azure 凭证、K8s Secrets、数据库密码、加密货币钱包、所有 .env 里的 API Key,全部被 AES-256 加密打包,POST 到攻击者的仿冒域名 models.litellm.cloud。如果检测到 K8s 环境,还会在每个节点部署特权 Pod 横向扩散。 最恐怖的是触发方式。攻击者在包里塞了一个 34KB 的 litellm_init.pth 文件。Python 的 .pth 是路径配置文件,由 site 模块在解释器启动时自动处理——如果某行以 import 开头,直接执行。攻击者利用这个机制写了一行: import os, subprocess, sys; subprocess.Popen([sys.executable, "-c", "import base64; exec(base64.b64decode('...'))"], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL) 不需要你 import litellm,不需要你调用任何函数。你跑 pip、跑 python -c、IDE 启动语言服务器,甚至 pytest 跑测试——只要 Python 解释器启动,恶意代码就执行。装上就中招,完全静默。 载荷是三层 base64 嵌套:第一层 .pth 启动子进程;第二层是编排器,内嵌攻击者的 4096 位 RSA 公钥;第三层是凭证收割器,系统性搜刮 /home、/opt、/srv、/var/www、/app、/data、/tmp 下所有敏感文件。收集完毕后用 openssl 生成随机 32 字节 AES 会话密钥加密数据,再用 RSA-OAEP 加密会话密钥,打包为 tpcp.tar.gz 外传。 收割器之外还有持久化后门:在 ~/.config/sysmon/sysmon.py 注册为 systemd 用户服务,每 50 分钟轮询 checkmarx.zone 获取新指令,下载到 /tmp/pglog 执行。启动有 5 分钟延迟来躲避沙箱分析。即使你卸载了 litellm,后门仍然存活。 而且 pip install --require-hashes 也拦不住——恶意文件正常列入 wheel 的 RECORD,哈希完全匹配,因为包本身就是用被盗的合法 PyPI 令牌发布的。 你可能从没手动装过 litellm,但 DSPy、MLflow、Open Interpreter 等 2000 多个包都把它当依赖。Mandiant 确认已有 1000+ SaaS 环境感染,预计扩展到 10,000。 而这次攻击差点完美得逞——唯一的破绽是攻击者自己代码有 bug。.pth 通过 subprocess.Popen 启动子进程,子进程初始化时 site 模块又扫描到同一个 .pth,再次触发,指数级递归形成 fork bomb,撑爆了一个 Cursor 用户的内存才被发现。Karpathy 说:如果攻击者代码写得再好一点,这件事可能几周都不会被发现。 更荒诞的是攻击链的起点:安全扫描工具 Trivy 在 3/19 先被攻陷,攻击组织 TeamPCP 用它窃取了 litellm 的 PyPI 发布令牌,3/24 直接往 PyPI 推送带毒版本。用来保护你的工具,变成了攻击你的入口。社区在 GitHub 提 issue 报告后,攻击者 102 秒内用 73 个被盗账号发了 88 条垃圾评论淹没讨论,然后用被盗的维护者账号关闭了 issue。 自查脚本(覆盖版本检查、.pth 搜索、后门检测、可疑连接、K8s 扫描): gist.github.com/sorrycc/30a7… 安全版本:litellm==1.82.6。装了 1.82.7 或 1.82.8 的,假设所有凭证已泄露,立即轮换。
42
173
20
857
175,002
i changed all our "loading..." states to "thinking.." we are an agentic AI startup now
187
590
117
9,926
422,243
OMG??? Really??? curiosity
What if Claws built a company? Over the past two weeks I ran an experiment with OpenClaw. My AI agents worked ~20,000 minutes: • writing code • fixing bugs • running marketing and sales We started joking it was a “claws company”. That experiment became a new project: Buda.
1
17
Refactored the architecture of Social-Agent S: it runs engagement loops across multiple platforms. P: adding new features was becoming painful. A: redesigned it into a modular, state-driven pipeline. R: now new platforms and behaviors can be plugged in much more easily.
1
36
Just added a Gateway feature to MCPSDK. You can now quickly integrate MCP Servers into IDEs like VSCode, Cursor, and Claude via local STDIO or remote HTTP. Makes connecting tools to MCP way simpler for developers. Still iterating, but this already feels pretty nice.
2
50
Most "One Person Company" guides focus on auto-writing and auto-posting across platforms. But after running a Social Agent for a while, I realized something: content generation is only half of the loop.
1
1
26
The harder part is distribution through interaction — browsing, liking, and commenting to surface your presence. Turns out growth looks less like publishing… and more like participating.
1
23
Models are becoming a commodity. The real moat is the proprietary data that powers them.
1
18
We are moving from coding software to growing hardware. The boundary between tool and agent is gone.
1
16
Benchmarks measure performance. Epiplexity measures potential. One is for the leaderboard, the other is for the product.
2
36
The AI Czar title is the prize. The hidden cost is becoming the firm's permanent tech support.
1
13
Vibing is for consumers. Mastery is for those who intend to own the infrastructure.
1
11