Andrew Hoh (@andrewwhoh) profile picture
Joined August 2023
Photo by Andrew Hoh (@andrewwhoh)Photo by Andrew Hoh (@andrewwhoh)Photo by Andrew Hoh (@andrewwhoh)Photo by Andrew Hoh (@andrewwhoh)
I built semantic search for the official MCP registry. Available both as a REST API and a remote MCP server. ETL updates it nightly. Hope it's useful for @MCP_Community
2
5
20
1,109
Andrew Hoh retweeted
Introducing the 2025 Exceptional 100 Recognizing today's most exceptional engineers, designers, researchers, and builders working within hyper-growth, high-velocity orgs. Brought to you by @ExceptionalCap @marell @TheNestVC @grahaminhoo @fresco_inferno
2
4
1
8
808
Introducing mcp-agent cloud (beta): an agent runtime exposed entirely over MCP. Deploy agents as remote MCP servers, and run them as durable @temporalio workflows. ▶️uv add mcp-agent ▶️mcp-agent init ▶️mcp-agent deploy Looking for early adopters to take this for a spin 🧵
5
7
3
23
2,391
Story behild the hack ultrathink.nyc/p/why-i-quit-…
MCP is now the de facto standard for LLM agents 200 developers came to NYC to compete for $10,000+ and prove what’s possible. Here’s what we saw at the Build the Future MCP Hackathon (🧵):
3
3
1
26
5,790
MCP is now the de facto standard for LLM agents 200 developers came to NYC to compete for $10,000+ and prove what’s possible. Here’s what we saw at the Build the Future MCP Hackathon (🧵):
45
216
29
2,724
713,484
Andrew Hoh retweeted
AI generated music is unreal!!! @elevenlabs
1
2
5
465
Andrew Hoh retweeted
Upgrade your mcp agents with @OpenAI's new open source models - gpt-oss-20b & gpt-oss-120b. Example code -> github.com/lastmile-ai/mcp-a…
1
3
1
9
1,467
OpenAI getting back into the open source game - openai.com/open-models/
22
Andrew Hoh retweeted
Our AI Engineer World's Fair talk just dropped! @qadri_sarmad explores a game-changing idea: what if agents were MCP servers? We're building this with mcp-agent at @LastMileAI, turning agents into composable, scalable building blocks that work across any MCP client. - Multi-agent orchestration - Plug-and-play workflows - Infra-grade scalability A paradigm shift toward more flexible, interoperable agent systems. Watch: youtu.be/uFPAtKIN-FQ?feature… Repo: github.com/lastmile-ai/mcp-a… Thanks @swyx @AI_Engineer! 🙏
3
6
222
This is currently trending on the frontpage of HN! If this is your first time seeing mcp-agent, you can check it out here: github.com/lastmile-ai/mcp-a…
2
4
214
How it works - Workflows: In mcp-agent, a Workflow is an agent application that uses LLMs and MCP servers to perform tasks (e.g. summarizing GitHub issues → Slack message, grading a student short story, etc.). The Agent Server exposes Workflows as MCP tools.
1
1
2
200
Agents as MCP Servers We've added a new paradigm to mcp-agent -- exposing agents as MCP servers themselves. Invoke agents directly from Claude, Cursor, or any MCP client—enabling multi-agent workflows, durable execution, and reusable agents. Details 👇
4
6
31
3,100
Andrew Hoh retweeted
LastMile AI will be at the #AIAgentConference 2025 in NYC 🗽 Our CEO @qadri_sarmad is speaking at 1:30 PM ET on building & evaluating production-ready agents. If you’re around, come say hi! 📍Jay Conference Center, Bryant Park
2
4
233
16/ mcp-agent is a simple, composable framework to build effective AI agents using Model Context Protocol and simple workflow patterns. @bllchmbrs
 teapaw.org/bllchmbrs/status/19101…
mcp-agent is a simple, composable framework to build effective AI agents using Model Context Protocol and simple workflow patterns. Key features: - Handles the mechanics of connecting to MCP servers and working with LLMs, allowing developers to focus on core business logic - Implements well-defined workflow patterns from Anthropic's "Building Effective Agents" in a composable way - Supports multi-agent orchestration using the OpenAI Swarm pattern - Enables programmatic control flow and seamless integration with human input and signals - Provides interoperability with any tool exposed by MCP servers
1
3
3
760
Andrew Hoh retweeted
MCP-Agent:基于模型上下文协议的智能体构建框架 @LastMile 核心概念 MCP-Agent 融合了两个 AI 开发领域的基础性进展: · MCP:由 Anthropic 推出的标准化接口,使任何软件都能通过 MCP 服务器被 AI 助手访问和使用 · 构建有效智能体模式:一套用于构建生产级 AI 智能体的简单、可组合的模式集合 主要优势 · 连接管理:自动处理 MCP 服务器连接的生命周期管理 · 模式实现:实现了所有"构建有效智能体"中描述的模式,并支持链式组合 · 模型无关:实现了 OpenAI 的 Swarm 模式,但以与模型无关的方式 关键组件 · MCPApp:全局状态和应用配置管理 · MCP 服务器管理:轻松连接到 MCP 服务器的工具 · Agent:有权访问一组 MCP 服务器并将它们作为工具调用暴露给 LLM 的实体 · AugmentedLLM:由 MCP 服务器提供的工具增强的语言模型 支持的工作流模式 · 并行(Parallel):将任务分发给多个子智能体,然后合并结果 · 路由(Router):根据输入,路由到最相关的处理类别 · 意图分类(IntentClassifier):识别最符合给定输入的意图 · 评估-优化(Evaluator-Optimizer):一个 LLM 优化响应,另一个评估它直到达到质量标准 · 编排-工作者(Orchestrator-workers):高级 LLM 生成计划,分配给子智能体,并合成结果 · 群体(Swarm):与 OpenAI 的实验性多智能体模式兼容的实现 应用案例 · Claude 桌面应用集成:将 MCP-Agent 应用集成到像 Claude Desktop 这样的 MCP 客户端 · Streamlit 部署:构建具有 Web 界面的 MCP-Agent 应用 · Gmail 智能体:使用文本提示对 Gmail 执行读写操作 · RAG 聊天机器人:使用向量数据库进行文本问答 · Marimo 集成:在响应式 Python 笔记本中实现智能体 · 客户服务智能体群:使用 Swarm 工作流处理不同类型的客户服务请求 未来路线图 · 持久执行:允许工作流暂停/恢复并序列化状态 · 记忆功能:添加长期记忆支持 · 流式处理:支持迭代进度的流式监听器 · 更多 MCP 功能:扩展工具调用以支持资源、提示和通知
build ai agents using Model Context Protocol and workflowpatterns
7
40
113
13,789
Andrew Hoh retweeted
build ai agents using Model Context Protocol and workflowpatterns
4
24
5
236
33,166
Browsing the Internet like an RPG game (MCP Servers: Puppeteer; Framework: MCP-Agent)
1
2
5
284
Andrew Hoh retweeted
🌐 LangGraph's Adaptive RAG using LastMile AI's AutoEval models - an upgraded way to self-correct LLM responses. Adaptive RAG is a powerful strategy designed to use self-reflection to refine and correct its responses. Adaptive RAG is a powerful strategy designed to use self-reflection to refine and correct its responses. @langchain 's tutorial demonstrates how to implement an Adaptive RAG application using LangGraph. We've taken it one step further by swapping out the LLM graders with our AutoEval models for Relevancy and Faithfulness. Instead of binary llm graders (yes or no), you can now configure thresholds to enforce stricter or looser graders within your graphs. Check out the cookbook to give it a try 🧐: github.com/lastmile-ai/lastm…
1
3
206