←
David Soria Parra3,971 posts
David Soria Parra avatar

David Soria Parra

@dsp_

Member of Technical Staff @AnthropicAI. Co-Creator of modelcontextprotocol.io. Ex-Meta. Playing with computers and tech

◷ Joined August 2008▧ 109 Photos and videos
Matt Ambrogi avatar
⟳ David Soria Parra retweeted
Matt Ambrogi@matt_ambrogi·22h

Lot of people asked about this. You can basically make your MCP self improving by doing this: - Add a submit_feedback tool to your MCP - Agent using your MCP will write detailed issue including proposed fix. - Have it send to an internal Slack channel - Have Claude in Slack channel and tag it to put up a PR when you agree w proposed change.

Post attachment

This is why you should actively invest in an embedded agent and an MCP. MCP observability only allows you to see tools called and you have to reverse engineer intent. If you have an actively used embedded agent, you would have seen this query and been able to ID that the proper filters were not in place. Fix the tool and it benefits both surfaces. You can also get a lot of benefits by adding a submit_feedback tool to your MCP to gather this data.

Nicolas Bustamante avatar
⟳ David Soria Parra retweeted
Nicolas Bustamante@nicbstme·Nov 14

I came to the conclusion that most SaaS should become an MCP server. we’re getting the economics of AI software backwards. A lot of AI startups are building their own chat interfaces, agents, and workflows, essentially reselling tokens from the big labs at low to negative gross margins. What if SaaS companies did the opposite? Take HR software. Instead of building an AI assistant, become the best MCP server for HR. Expose tools to: - Benchmark salaries against proprietary market data - Run compensation reviews and calculate pay equity - Model salary increases against a department’s budget - Recommend promotions based on company policies - Update employee compensation in the HRIS - Manage approvals, permissions, and audit trails Let ChatGPT, Claude, or Copilot handle the conversation, reasoning, and orchestration. Let the customer pay for the tokens. The SaaS company keeps the valuable stuff: proprietary data, integrations, business logic, permissions, compliance, and the system of record. You still charge a SaaS subscription, but you’re no longer paying to generate every answer. You can ship faster, with fewer engineers, and potentially restore software-like gross margins. The next generation of SaaS might not have much of a UI at all. Just an incredibly good MCP server.

ClaudeDevs avatar
⟳ David Soria Parra retweeted
ClaudeDevs@ClaudeDevs·Nov 12

Self-hosted Claude Managed Agents is now in private beta. You can host the agent harness, sandboxes, and session history, in your own cloud or on-premises, and choose where requests to Claude are sent.

Post attachment
David Soria Parra avatar
David Soria Parra@dsp_·Nov 12

I think this is pretty cool. Check it out. Also props to Samuel for the epic video.

🚀 Today we're launching Agent Sandboxes in @pydantic Logfire: a profoundly different kind of sandbox. "The world doesn't need another sandbox" they say. Maybe. Or maybe the world doesn't need another firecracker/gVisor/VM infra-fest. Our Agent Sandboxes are different. They're just an isolated interpreter with cold start times measured in milliseconds, where the code can only call the tools you give it and use the directories you mount. A massive bonus is you can snapshot and resume a REPL weeks later, with the snapshots costing virtually nothing. Programmatic tool calling, solved. Start free, $1000 of free credit to try it in the first month 👇 pydantic.io/launch-agent-san…

1:32
David Soria Parra avatar
David Soria Parra@dsp_·Nov 11

True acts of love

Was struggling with some game design UX so asked my girlfriend (a designer) if I could borrow her design skills and she literally emailed me a SKILL.md 😂

Post attachment
Blendi avatar
⟳ David Soria Parra retweeted
Blendi@BlendiByl·Nov 11

i used fable 5.2 to make this "hand-painted" game in less than an hour Every frame is about 1.5 million brush strokes, running at 200 fps in a browser tab. Open source too!

0:16
Liad Yosef avatar
⟳ David Soria Parra retweeted
Liad Yosef@liadyosef·Nov 10

...and coming soon to the MCP Apps spec itself! Entry points are a great example of a feature that started as a specific host implementation and made its way upstream into the official spec. Write once, run everywhere 🚀

powered by OpenAI plugin extensions 🤯 github.com/openai/mcp-extens…

Post attachment
Claude avatar
⟳ David Soria Parra retweeted
Claude@claudeai·Nov 10

Also new: with interactive connectors, you can now open the full version of a tool inside Claude. Work in Adobe, Atlassian, and Box beside your conversation. Make changes yourself, or ask Claude to make them with you. Rolling out on web and desktop.

1:00
David Soria Parra avatar
David Soria Parra@dsp_·Nov 10

It is pretty obvious to me that AI enables a new era of formal verification. On MCP we just working with some people to formally verify the SDKs and it already yields results.

I had Opus 5.5 formalize CPython in Lean: 613,318 lines of Lean that copy Python's internals line by line, tested against the real interpreter and checked by Lean. On the way it found real bugs in Python, including crashes from plain Python code. 🧵

0:11
Felix Rieseberg avatar
⟳ David Soria Parra retweeted
Felix Rieseberg@felixrieseberg·Nov 9

I wanted to test Fable 5.2's quality with one-shot prompts without human feedback. Everything in this game is from scratch: models, sounds, textures. ~$550 at API pricing, ~2.7M output and 445M input tokens, heavily cached. Took 8 hours! Play it here: claude.ai/artifact/BGY1SztrA…

1:12
David Soria Parra avatar
David Soria Parra@dsp_·Nov 7

Yes this. It is so good that some apps like Fusion 360s and others are shipping like that and I would love to see more of it. Having Claude connect to your local app and seeing the changes is the best of both worlds. Agent interaction + native app

a pattern I’m liking: creator apps like Instagram Edits and Roblox Studio shipping with built-in MCPs, so agents can do some of the work while power users can collaborate and continue the work in tools that they're used to

Tyler Brown avatar
⟳ David Soria Parra retweeted
Tyler Brown@TylerBrownAT·Nov 6

Craziest engineering-to-impact ratio I've seen in my career: our MCP. Platform teams power every surface behind the scenes of course, but on the product side, it's amazing how few engineers support something this heavily used. And growth keeps accelerating past what I thought was possible. Seriously. The past month has been crazy. I think the reason is simple: agent experience is easier than user experience. It just is. State is the root of so many bugs, and frontends are full of it. An MCP has far less, so you ship way faster. And of course, it helps to have an amazing MCP platform team behind us.

Nils Adermann avatar
⟳ David Soria Parra retweeted
Nils Adermann@naderman·Nov 5

Just set this up for a couple services, feels like magic 😯

This is a big step towards an truly seamless experience for users. If you build for Claude make sure you support this.