Today, we’re excited to present MCPJam’s ngrok tunneling feature, enabling you to observe all JSON-RPC messages sent between ChatGPT and your app via the ngrok tunnel. Instantly ngrok your tunnel in the inspector. 1️⃣ Observe the communication layer between ChatGPT and your app’s

We wrote a blog post detailing how MCP server instructions are used and included a template to get started. 🔗: https://t.co/tWJsg5qUMl 🧵 (3/3)

What MCP server instructions should do: - Keep it concise and scannable - Use clear section headers - Explain patterns and conventions - Document dependencies between features - Note performance/timing expectations - Include practical usage hints DO NOT DO: - Duplicate tool

MCP server instructions is an overlooked feature Server instructions in the MCP spec is a dedicated mechanism for servers to provide LLMs with essential contextual knowledge like tool interdependencies and operational constraints. Let's go over some of the best practices: 🧵

Building a well designed MCP server is an art. In the MCP community, it’s increasingly described as “tool ergonomics”, designing MCP tools so that an LLM understands how to use them. Running evals is the most effective way to measure, iterate, and improve these ergonomics.

Context7 documentation is now live 🚀 Powered by @mintlify

Been playing with MCPJam, easily the best visual MCP testing tool I’ve tried! Definitely got some feedback like I wish I could see a formatted version of the responses, but overall kudos to @mcpjams for the great work.

Debugging MCP authorization no longer sucks We built an improved OAuth debugger in the inspector that lets you see what happens at every step of the handshake. This helps with pinpointing exactly where the issues are in your auth implementation. New features include: 1️⃣

MCP authorization can be pretty intimidating We put together a MCP authorization checklist with the draft November spec that shows you exactly what’s happening at every step of the auth flow, with code examples. 🔗 Blog:

MCPJamming on a train with @Starlink SF ➡️ Chicago

What’s new in Playground-V2: 1️⃣ Render MCP-UI and OpenAI apps SDK. We have support for servers built with MCP-UI and OpenAI apps SDK. 2️⃣ View all JSON-RPC messages sent back and forth between the MCPJam client and MCP server for fine debugging. 3️⃣ Added free frontier models

The MCP playground has been the most essential part of the project. With growing interest in MCP-UI and OpenAI apps, we’re doubling down on the playground. I’m excited to release our new playground - Playground V2. https://t.co/eRnoCTtrb2 #mcp

We can also unit test MCP resources, prompts, disconnects, and more. I wrote a blog article on MCP unit testing here: 🔗

Unit testing MCP servers is incredibly simple: We've set up unit tests with Jest and 𝐌𝐂𝐏𝐂𝐥𝐢𝐞𝐧𝐭𝐌𝐚𝐧𝐚𝐠𝐞𝐫, the first addition of our mcp/sdk. Here are some components of the MCP server we can unit test. 1️⃣ Server connections - client connects to the server, test

We've been rebuilding the MCP Playground from scratch with @aisdk. So clean. Front end - React app with useChat hook Hono backend - Hono server with toUIMessageStreamResponse Been experimenting with our MCPClientManager too. Wish we had this implementation earlier.

This unlocks so much for OpenAI app development

Everyone should be able to use frontier models to build. Starting today, GPT-5, Sonnet 4.5, and more will be free of use at MCPJam. @MCP_Community

Model’s we’re releasing: - Claude Sonnet 4.5, Haiku 4.5, - Gemini 2.5 Flash, Gemini 2.5 Flash Preview, - GPT-5 Codex, GPT-5 Mini, - Grok 4 Fast, - Kimi K2, - GLM 4.6 My goal's always been to help people build better MCP servers. As a server developer, you have to consider

We might be setting money on fire, but 100% it's worth it. I'm excited to announce that we're providing frontier proprietary/open source models from OpenAI, Anthropic, Gemini, and more to be used for free in MCPJam's LLM playground. You no longer have to bring your own API key

Building an MCPClient that can replace @aisdk's experimental_createMCPClient We built a MCPClientManager with an adapter for Vercel AI SDK. Our goal's to standardize how we build MCP clients. MCPClientManager was proposed to be a part of the official Typescript SDK