Mike Coughlin

The top MCP question from enterprise leaders right now: "How do I curate an approved list of servers for my org?" With 13,000+ MCP servers out there and new ones shipping daily, "just let people install things" isn't a strategy. So we teamed up with @tadasayy and @grumpygrowthguy from @pulsemcp to build a curated catalog of trusted MCP servers purpose-built for Speakeasy customers rolling out MCP at scale. Production-grade, vetted, enterprise-ready. 🤝

This is really neat but it’s not a design tool as much as it’s a design _production_ tool. The practice of design is mostly about what comes before production. There’s no doubt in my mind that all parts of software production will become automated very soon. Writing code, making web pages, putting pieces of a design system together etc. And that’s fine. I think few people actually enjoy this kind of production work. Wouldn’t it be better if we spent our precious time in life on what is more meaningful?! At the core, the practice of design is methodical; like architecture, not like art. In a nutshell: We find constraints, form comprehension of the whole and propose solutions that honor those constraints. First after that do we enter some form of production phase, usually prototypes first, learn about some constraints that were hidden before, loop back, prototype and then build the production-grade “final” artifact. These last few tasks are quickly losing value because AI tools can do it much faster (not yet better though) than humans. It’s simply just what has the best RoI for a business. Some companies and individuals will continue to spend human time on certain parts of the “production line” as a market differentiator, but it will cost them a relatively high price compared to competitors. Anyhow, I still haven’t seen a tool better than Figma that supports the actually-interesting part of the design process. I wouldn’t be surprised if Figma focused their products on that, maybe separating “products for production” of “products for ideation & exploration.” The latter would obviously still leverage AI, but not to do the work for me but rather to support my efforts the way a therapist helps me live a better life (not living my life for me.)

@zeeg @trq212 Come on! Don't you have 20 Claws checking your calendar so you know when to wipe your ass?

@trq212 i never run out can you tell me what ridiculous activities i should be doing so that i can?

I want to do a few more of these calls. If your MAX 20x plan ran out of tokens unexpectedly early and you're willing to screenshare and run some prompts through Claude Code please comment. Trying to figure out how we can improve /usage to give more info.

Man, ngl, this is a rollercoaster ride.

@bcherny Hey, thank you! As a Claude Code user and an API user that wants to see Anthropic be more generous and transparent with usage limits, it sucked seeing some slop cannons abuse subscriptions and ruin it for the rest of us.

As we all work to push AI forward, I keep thinking of this awesome teaser for Interstellar: youtube.com/watch?v=nyc6RJ…

Agents, meet the Figma canvas

Now you can use AI agents to design directly on the Figma canvas, with our new use_figma MCP tool and skills to teach them. Open beta starts today.

Claude Cowork works because it’s connects to your services via MCP. Your Slack Claude Code plugin works because of MCP. A common auth standard is hugely valuable. If you have context bloat, get a better client, because you shouldn’t just dump a list of tools into your context window.

Spicy CFO of @figma, the Dalai Lama of CFOs. The man doesn't do earnings calls, he holds court.

Very cool, @figma!

Software engineers worried about AI absorbing their craft would do well to embrace the change and do what the history of building software has always done: move up the abstraction layer. Agentic engineering is here. Teams are making it work. The portfolio of very public examples of this adoption grows by the day. I’ve been coaching some of the teams going through this shift in the software engineering trade - and I’ve found that the mindset shift often takes just two very specific “aha” moments.

When we ask an agent to "fix this bug," we're approving hundreds of actions we'll never see. These agents have our production credentials; and increasingly, they run in the background while we work on something else. Engineers are spinning up multiple Claude Code sessions in parallel, and ClawdBot hit >100k GitHub stars in weeks by letting agents handle email, workflows, even car purchases. The shift to agents without constant supervision is here. Today we're launching @MintMCP_AI : governance for AI agents.

@techgirl1908 Keep preaching the good word, @techgirl1908

I see ppl try this then end up circling back and adding an MCP to the Skill. Most agents have built-in shell timeouts. So long running commands will timeout whereas MCP keeps the connection alive indefinitely. Also, MCP gives you streaming output, structured responses, caching, error handling... CLI gives you stdout parsing and exit codes 🙃

Giving away this shirt to a random person who retweets. #ProBowlVote Ben Skowronek #ProBowlVote @BennySkow

@voxmatt , CEO of @getclockwise , takes us through the difference between data access and intelligent coordination when building MCP servers. Most MCP servers expose calendar data for LLMs to process, Clockwise's server does the scheduling optimization server-side and returns actionable proposals. One leverages the LLM's intelligence. The other leverages domain intelligence. Read it here: pulsemcp.com/posts/how-we-b…

There are several kinds of MCP servers out there: - Thin wrapper logic that could be an Agent Skill or a CLI tool - CRUD access that does little more than use MCP as a standardized auth mechanism - And the most compelling: buckets of hard-won domain intelligence That third category is where MCP really shines

Happy birthday MCP! ❤️

MCP gateways have proven to be a critical piece of infrastructure of bringing MCP into enterprise settings, and that gateway infra is enabling creative ways of solving downstream MCP challenges. One of them: solving tool overload on a use-case by use-case basis.

I just watched @theo new video on why MCP sucks. So I thought I would explain why MCP actually doesn't suck. Protocols only exist to connect two distinct systems in a standard way. An LLM can only perform actions through tool calling. Tool calling is the standard interface. If I'm writing an application that uses an LLM and wishes to do tool calling, I don't need MCP. The code that calls the LLM can just implement the tool calling interface and be done with it. This is essentially theos use case and probably why he doesnt articulate MCP well. If you want the tool implementation to be independent of the code calling an LLM this is where a protocol comes in. Prior to MCP most approaches to independently distribute tools focus on language frameworks like langchain. No language independent way existed such that a LLM client in python could connect to a say a Java tool. But furthermore no standard way existed such that one could expose a tool publicly in a secure multi-tenant fashion. MCP solves this problem. And despite what @theo stated, MCP does define auth, it's one of the core features of the spec (yay OAuth 2.1, DCR,and DCIM 😭) If you use MCP for 1 tool or 1000 tools, it still has value. The issue most people complain about today is that there are too many tools or poorly defined tools. This is an issue independent of MCP and where a lot of exciting work like code mode is being done to address it. But there's other progressive discovery methods that exist, such as the implementations of ChatGPT apps, which is based on MCP, or Claude skills. Code mode, or progressive discovery don't remove the need for MCP, they just provide a meta approach to deal with tool scalability issues but can still be implemented with MCP. One could possibly make the argument that we just don't need to ship tools as systems of their own, but people are doing it and getting value out of it, so 🤷‍♂️