🛠️ Building with AI and Agents

​From 'no-reply' to 'please-reply': How to build an AI-Powered Email Assistant with Langflow and SendGrid​​

Tejas Kumar​
The internet is full of articles going back years telling people that sending emails from a "noreply" email address is terrible for your business and users. However, receiving and responding to emails from your users is time consuming and costly. What if you could just get an agent to reply? Tejas dives into how to set this up using Twilio SendGrid and Langflow.

We actually sent this email from please-reply@langflow.org, so you can just reply and ask anything about Langflow.

​Multimodal Support in Chrome's Built-in AI​

Raymond Camden​
At Google I/O in May, Google announced that they were experimenting with a multimodal language model built into Chrome Canary. Raymond shows how to use the prompt API to analyze images directly in the browser. Local models cost developers less to build with and most importantly preserve user privacy. Check out what you can achieve!

​How we built our multi-agent research system​

Anthropic
​This article contains many, many lessons from building multi-agent systems at Anthropic.

⚡️ Everything MCP

​What's new with MCP​

A big update was released to the protocol last week, including a focus on security. MCP, like everything else in generative AI, is very new and changes are to be expected, so keep an eye on these updates.

​Building an MCP Server with Nx​

Max Kless
​This article doesn't just show you how to build an MCP server, but how to do so in an Nx monorepo, debug it with the MCP inspector, and release it as an executable package to npm. Comprehensive.

​The remote GitHub MCP Server is now in public preview​

There is going to be a trend of MCP servers moving from local to remote. The GitHub remote MCP server runs the same code as the open-source, local version, you just don't need to run it yourself.

🗞️ Other news

​The lethal trifecta for AI agents: private data, untrusted content, and external communication​

Simon Willison
​If you've seen mentions of MCP vulnerabilities recently, Simon explains why they exist. It's really just a form of prompt injection, which remains hard to circumvent. There are some ideas, but this is important to think about as both a user and builder of these systems.

​Check out the videos from LangChain Interrupt​

The line-up is just full of people building agents and sharing their experiences. You can catch up on the videos now.

🧑‍💻 Code & Libraries

• ​OpenAI released their Agents SDK for TypeScript
  This joins the existing Python SDK. Check out the TS SDK documentation here.​
• ​agents-towards-production is a new set of tutorials for building agents, from Nir Diamant, the creator of the popular RAG_Techniques and GenAI_Agents repos. Lots to dig into.
• ​DSPy v3 beta1 is released​
• ​Hono, a web standards based web server, has released a middleware for building or adding MCP over Streamable HTTP Transport to your app.

🔦 Langflow Spotlight

Each newsletter we like to share a Langflow component that you can use to create your AI flows. As this is the first iteration of our new newsletter, there can be no other component to share than the Agent component.

The Agent component is the centerpiece to so many flows, coordinating tools and other agents seamlessly. Just pick your model provider and model, hook up a few components as tools and you're away. You can even make agents, or entire flows, into tools that an Agent component can use.