Matt Wang

@nicoalbanese10 @aisdk Nice add!! Would you consider having AI-SDK provide the full agent harness like Claude SDK and Cursor agent SDK has?

Wild that it only took a few prompts to build an @aisdk community provider for this!

I spoke at @mcpsummit in NY on "Evaluate What You Can't See." The core problem I discussed: your MCP server can execute perfectly and still fail the user. You have almost no visibility into it. Covered the User-Value Chain framework, why REST instincts don't transfer to MCP tool design, and 6 concrete ways to get user context in a system that hides it from you. Full 30-min talk: youtu.be/XqjOcKEUSRQ?si… #mcp #modelcontextprotocol @mcpjams

Your users aren't in your app anymore. They're talking to an agent in ChatGPT, Claude, or Cursor. That agent decides when to call you. It has all the user's context. You get almost none of it. Shipping an MCP integration isn't enough. You have to know if it actually delivers user value. mcpjam.com/blog/effective…

@mcpjams might be the most under-appreciated devtool that I can think about... so many things that would be an absolute pain to wrangle and debug is made simple

@rohitiwnl Appreciate the shoutout!!

@fashiongiik @GithubProjects Thank you Hanh!!

@GithubProjects MCPJam is great! We used MCPJam in usefractal.dev

MCPJam is the local inspector for MCP apps and ChatGPT apps. Build and test locally with: • Full widget emulator. • MCP server inspection. • OAuth debugger. • LLM playground. Iterate on your MCP apps faster than testing directly in ChatGPT or Claude.

@excalidraw @AnthropicAI @claudeai My favorite MCP app so far. Congrats on getting that out!

Thanks to good people at @AnthropicAI we now have an official MCP for Excalidraw! Take it for a spin on @claudeai (search for Excalidraw in Connectors, or use in Claude Code and elsewhere). More to come. ✌

Full launch blog: 🔗: mcpjam.com/blog/views

You can now share and mock MCP app views. We're excited to launch Views in MCPJam, you can now save your MCP app UI iterations, effectively taking a screenshot of your UI in that moment. With Views, you can: 1️⃣ Save views to track your app's UI progress over time 2️⃣ Share different UI drafts with teammates 3️⃣ Mock data to see what the UI would look like in different states The HTML and hydration data is saved, allowing Views work without having to be connected to your MCP app. Views are on the latest version of MCPJam. Full launch article in the thread 𝐧𝐩𝐱 @𝐦𝐜𝐩𝐣𝐚𝐦/𝐢𝐧𝐬𝐩𝐞𝐜𝐭𝐨𝐫@𝐥𝐚𝐭𝐞𝐬𝐭

Text streaming is now supported for free models in MCPJam. Excited to announce that we now have streaming for the free models we provide in the LLM playground and Apps Builder. No need to bring your own API key, it's on us. Start e2e testing your MCP servers and UI apps in the playground for free, on the latest version of MCPJam: 𝐧𝐩𝐱 @𝐦𝐜𝐩𝐣𝐚𝐦/𝐢𝐧𝐬𝐩𝐞𝐜𝐭𝐨𝐫@𝐥𝐚𝐭𝐞𝐬𝐭

I got the chance to write a guest post on @OpenAIDevs Blog about 15 lessons learned building ChatGPT Apps. developers.openai.com/blog/15-lesson… If you're building Apps, I think it's worth a read!

Case study 🔗: mcpjam.com/blog/asana-mcp…

@asana uses MCPJam to build ChatGPT apps and Claude Connectors. Recently, the Asana team launched their ChatGPT app and Claude Connector, bringing the Asana UI experience directly into ChatGPT and Claude for hundreds of millions of users. Asana engineers use MCPJam to develop and iterate locally, instead of having to redeploy and spin up ngrok on every change. Engineers can see code changes reflected immediately, reducing the feedback loop from minutes to seconds. MCPJam helped the Asana team move quickly from ideation to a production-ready app in just over a month. Check out the case study in the thread

Article: mcpjam.com/blog/mcp-apps-…

We teamed up with the team at @goose_oss to put together 5 tips for building great MCP Apps (MCP-UI) 1️⃣ Adapt to the host environment using hostContext 2️⃣ Control what the model vs UI sees 3️⃣ Handle loading and error states when listening to tool results 4️⃣ Keep the model in the loop by updating model context 5️⃣ Control tool visibility Wrote an article on how to implement each of these tips, with code snippets. Link to the article in the thread below! 👇

@goose_oss Thanks for jamming on the article with us!!

We teamed up with @MCPJams to share everything we learned building MCP Apps. Here's some tips that would've saved us hours: → Adapt to host themes → Control what the model sees → Handle loading states properly → Keep the model in sync → Restrict who can trigger tools Link to blog in the thread 👇

@kamathematic @gregpr07 That’s a crazy find 😂

@gregpr07 @mcpjams spotted 👀

wtf, adding skills with npx skill add creates all these folders in a local project?

MCP App / server builders should start thinking about writing skills to pair. We launched two features to help test Skills + MCP pairing: 1️⃣ View your skills beautifully in the skills tab. MCPJam lets you upload skills directly, which are then saved to your local skills directory. 2️⃣ Test Skills + MCP pairing in the LLM playground. Inspect how different skills improve your MCP server use Skills + MCP is a new and growing concept, would love to hear your opinions on it. Full article here: mcpjam.com/blog/skills

You can now test Skills + MCP pairing on MCPJam There’s a growing trend of using skills paired with MCP servers. MCP provides the tools, skills that teach the agent how to use it and add additional context. We’re also seeing @AnthropicAI encourage the use of Skills + MCP in their products. Anthropic recently launched the connectors marketplace and encourage app devs to submit skills to compliment the app.