Dec 28, 2025 · 1h 39m · latent-space
One Year of MCP — with David Soria Parria and AAIF leads from OpenAI, Goose, Linux Foundation
David Soria Parra · 51m spoken Jim Zemlin · 13m spoken Nick Cooper / Brad · 10m spoken Shawn Wang · 7m spoken Alessio Fanelli · 6m spoken
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this special Latent Space episode, co-hosts Alessio Fanelli and Swyx celebrate the one-year anniversary of the Model Context Protocol (MCP) with co-creator David Soria Parra and host a historic panel featuring leadership from OpenAI, Block, and the Linux Foundation to announce the launch, architecture, and governance of the Agentic AI Foundation (AAIF).
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 16.3% of the talking time here. How this is scored →
The hosts as informed peer 6.0 Guest teaching 5.5 Guest disagreement 1.5 The hosts pushing back 1.8
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 21:11 David rejects the 'MCP vs Code Mode' binary
David pushes back forcefully on the community misconception that code execution modes render MCP obsolete, clarifying that MCP remains the foundational communication and authentication layer.
Hardest push from the hosts ▶ 55:40 Swyx challenges iframe architecture for MCP appsSwyx directly challenges David on the viability of iframes for rich UI components, highlighting severe CORS headaches and UI styling inconsistencies.
Biggest teaching moment ▶ 8:59 OAuth architecture and IDP decoupling in enterpriseDavid provides a masterclass on enterprise authentication, walking through why combining the auth server and resource server broke enterprise SSO integration and required complete architectural separation.
The host holds their own ▶ 26:55 Swyx surfaces Anthropic's internal MCP gatewaySwyx demonstrates deep domain knowledge by citing internal engineering talks to drill into Anthropic's actual internal platform infrastructure and dogfooding practices.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Reflecting on One Year of the Model Context Protocol | 5 | 4 | 1 | 1 | Swyx and Alessio establish a collegial atmosphere, accurately framing the one-year milestone and public announcements. David gives a structured historical recap of the protocol's evolution from local to remote and enterprise auth. | |
| The Creation and Governance of the Agentic AI Foundation | 6 | 4 | 1 | 2 | Alessio and Swyx ask pointed questions about how other protocols and the foundation governance fit together, referencing Swyx's early analysis on MCP. David clarifies the narrow initial scope for protocol acceptance and core maintainer roles. | |
| Iterating on Remote Transports and Enterprise OAuth Specifications | 5 | 8 | 2 | 2 | David openly schools the hosts on OAuth architecture, breaking down why their initial spec failed by combining the auth server and resource server instead of integrating with enterprise IDPs. Swyx asks clarifying questions to understand the failure mode. | |
| Transport Protocols, Bidirectional Streaming, and Horizontal Scalability | 6 | 7 | 2 | 1 | Alessio asks about transport protocol design choices and regrets. David provides a deep technical breakdown of why streamable HTTP was chosen, why making bidirectional return streams optional backfired, and the distributed state challenges of horizontal scaling. | |
| Open Source Standards: IETF Comparison and Core Maintainer Governance | 6 | 5 | 2 | 2 | Swyx draws a comparison between the MCP governance model and the IETF standards process. David explains why IETF consensus moves too slowly for AI velocity and defends MCP's small core maintainer benevolent dictatorship. | |
| Model Alignment, Progressive Discovery, and Programmatic MCP vs Code Mode | 7 | 6 | 4 | 3 | Alessio and Swyx bring up context bloat and probe the tension between Anthropic's programmatic MCP and Code Mode. David firmly clarifies that programmatic MCP is an execution optimization rather than a replacement for protocol connectivity. | |
| MCP vs Skills: Vertical Domain Knowledge Meets Horizontal Connectivity | 6 | 6 | 2 | 2 | Swyx prompts a comparison between MCP and skills. David delineates them clearly, framing skills as vertical domain knowledge needing execution environments while MCP serves as the horizontal communication and auth layer. | |
| Anthropic's Internal Dogfooding and Enterprise Gateway Infrastructure | 7 | 4 | 1 | 1 | Swyx demonstrates deep familiarity with Anthropic's internal setup from prior talks, highlighting internal MCP gateways. David confirms and details how internal employees deploy containerized MCP servers autonomously. | |
| Standardizing MCP Registries, Discovery, and Supply Chain Security | 6 | 6 | 2 | 2 | Alessio queries how public and pass-through registries fit together. David explains the NPM-like trust model, the risk of supply chain attacks, and the architectural necessity of curated private sub-registries. | |
| Lessons from Developer Summits and Financial Services Industry Constraints | 5 | 7 | 1 | 1 | Swyx asks for field lessons from international developer events. David explains unique legal constraints in financial services, like mandatory third-party attribution contracts spearheaded by Bloomberg. | |
| The State and Evolution of the MCP Sampling Protocol | 7 | 6 | 2 | 2 | Swyx and Alessio press on why sampling adoption is lagging and how long-running tasks operate. David explains why clients avoided sampling and details the polling-and-webhook design of the new tasks primitive. | |
| Context Window Compaction, Retention Strategies, and Agent Memory | 6 | 6 | 3 | 2 | Swyx asks about context compaction and Alessio asks why developers should use MCP over standard REST APIs. David explains REST OpenAPI specification bloat and ambiguous tool descriptions. | |
| MCP Apps and UI: Bringing Rich Visual Experiences into Chat Interfaces | 8 | 5 | 2 | 6 | Swyx aggressively pushes back against using standard raw iframes for MCP apps due to CORS issues, styling fragmentation, and brand consistency concerns, prompting David to acknowledge styling remains an open problem. | |
| Protocol Extensions, Healthcare Compliance, and Call for Community Builders | 5 | 6 | 1 | 1 | Swyx asks about additional protocol extensions. David provides an example of session-level isolation for HIPAA compliance to prevent data leakage across servers. | |
| Neutrality and Industry Protection: Why Anthropic Donated MCP to the Linux Foundation | 6 | 5 | 2 | 2 | Swyx directly raises community skepticism about Anthropic taking its eye off the ball by donating MCP. David firmly refutes this, citing proprietary standard lock-in risks like HDMI on Linux. | |
| Introducing the Agentic AI Foundation Leadership Panel | 5 | 4 | 1 | 1 | The panel is introduced with Jim Zemlin, Nick Cooper, and Brad. Alessio asks about the origins of the multi-company coalition, and David and Nick detail the organic collaboration. | |
| The Linux Foundation Perspective: AI Dog Years and Unprecedented Demand | 5 | 6 | 1 | 1 | Jim Zemlin explains the Linux Foundation's criteria for launching new efforts, emphasizing the unprecedented inbound interest and describing how AI moves in 'dog years.' | |
| Goose as the Reference Agent Client for Open Protocol Development | 6 | 5 | 1 | 1 | Brad and Nick describe Goose's role as an early reference implementation client to validate protocol specs like MCP UI before standardization. | |
| Balancing Open Standards with Healthy Implementation Competition | 6 | 5 | 1 | 2 | Swyx asks whether OpenAI's open-source agent tooling might be donated. Nick explains the delicate balance between open standard commonality and competitive diversity in implementation. | |
| Incentives for Contributing Open Source Projects to the Foundation | 6 | 5 | 1 | 1 | Alessio asks why an independent project with strong traction should donate to the foundation. Jim, Nick, and Brad explain the positive feedback loop of pooled engineering resources and shared learning. | |
| Why Agents Warrant a Dedicated Foundation Distinct from LF Data & AI | 7 | 5 | 1 | 2 | Swyx queries why a distinct foundation was formed instead of folding agents into LF Data & AI. Jim explains the unique audience separation between model training infra, inferencing, and agent application layers. | |
| Inside the Directed Fund: Operations, IP Protection, and Community Infrastructure | 6 | 6 | 1 | 2 | Alessio and Swyx ask where directed funds actually go in practice. Jim details IP protection, community convening logistics, metrics platforms, and separating technical governance from financial contributions. | |
| Drawing Lessons from Kubernetes and the Cloud Native Computing Foundation | 6 | 6 | 1 | 2 | Alessio compares the current landscape to the CNCF era. Jim recounts the historical precedent of Google donating Kubernetes in 2014 when Mesos and PaaS were competing standards. | |
| 2025 Vision for the Agentic AI Foundation and Concluding Thoughts | 5 | 4 | 1 | 1 | The panel shares concluding thoughts and 2025 aspirations, highlighting asynchronous agent workflows and enterprise adoption milestones. |
Statements from this episode (45)
Altman, Nadella, and Pichai publicly committed to adopting MCP around April
“And then like, you had this like inflection point around April with like Sam Altman and Satya and Sundar and all posting about like MCP and that they're going to adopt MCP at Microsoft, at Google. At OpenAI and that was really like the big inflection point.”
Anthropic added long-running tasks to MCP for deep research and communication
“We just launched like in our end of like November, then the recent iteration of the protocol, finally, like the next bigger improvement to the protocol, which is like long running tasks to really allow for like, you know, deep research type of task and like, y…”
Agentic AI Foundation is the first open-source foundation founded by Anthropic
“Like for us, it's the first time we at Anthropic have an open source foundation.”
Other agent protocols currently lack the adoption required for AAIF inclusion
“At least on the protocol side, like a defector standards. And I don't think any of the other protocols, it feels like they're not just there yet, but of course, if they get there, then we are like super open as long as they're like complementary to what's, wha…”
MCP decoupled authentication from resource servers to support enterprise identity providers
“In the first iteration of our authentication respect, we combined them together into the MCP server... And if you combine these into the same server, you just can't do this anymore. And so all we needed to do is like, okay, we are a resource server. The MCP se…”
OAuth is fundamentally designed for humans, failing agent-to-agent authentication workflows
“OAuth itself is for the most part, a very human centric protocol. It's just, it just tells you how you obtain a token. If you don't have a token, once you have a token, actually it doesn't matter.”
MCP currently lacks authentication for mutually unknown clients and servers
“But if the client and the server don't know each other, we just don't have a good solution for now.”
MCP must enable inherently stateful agents with long-term client-server communication
“MCP should also enable agents, and agents are inherently somewhat stateful, and there's some form of, like, long-term communication going between, like, the client and the server, and so we always looked for something like that.”
Making return streams optional in Streamable HTTP was an MCP design mistake
“We got wrong that. We made a lot of things optional for the clients to do. Like you can, the client can connect and open this return stream from the server, but it doesn't have to. And the reality is, is no client does it because it's optional. And so a lot of…”
Google and Microsoft are already running MCP at millions of requests
“And we know from like some companies like the Googles of the world, the Microsoft of the world, they're doing MCP at a scale that I can't tell you the numbers, but it's like in the million of requests.”
MCP is governed by a core maintainer group of eight people
“And so that's why we run MCP as like a really traditional open source project with like a core maintainer group of like eight people that basically decide everything. And then like input from everybody else. Like we get input and people can make suggestions an…”
Underlying LLMs are technically completely uninvolved in the Model Context Protocol
“MCP is a protocol between the AI application and like servers, right? So the model is actually technically not involved in MCP.”
Most AI applications will eventually adopt sandboxed execution environments for code
“And I think you will see evolutions as we're getting better of like how we use these models and the infrastructure around it gets a bit more mature. And you suddenly can assume that most model like AI applications will have some form of like Sandboxing for exe…”
Skills supply vertical domain expertise while MCP provides horizontal system connectivity
“What skills do, they like, they give you the domain knowledge for like a specific Set of tasks, like how you are, how you behave, how should the model behave as a data scientist, or how should the model this behave as an accountant or whatever. But MCP gives y…”
Anthropic created MCP so rapidly expanding internal teams could build integrations independently
“MCP before we even open source it was born of the idea of like, I'm in a company that is growing crazy. I'm in the development side of things, development tooling side of things. I will grow slower than the rest. How can I build something that they can all bui…”
Enterprise MCP adoption is growing much faster internally than publicly visible
“In big enterprises, you see MCP everywhere. And it's actually way growing way faster than you would think because it's mostly internal to companies and without people seeing it.”
Enterprises will adopt private MCP registries similar to npm and PyPI
“They're effectively using private registries already for NPM. They're using for PyPy and they will also do it for MCP servers. In there you have implicit trust and then you can just search.”
Bloomberg is heading a financial services interest group for MCP standardization
“And so now that's when we, like, started, like, creating this financial services interest group that Bloomberg is heading up to, like, figure out, like, what are some of the things that you, like, a client must do if it wants to speak to our financial services…”
Anthropic plans to host an MCP developer summit in New York
“We're going to do it in New York in April, I think, or March or something like that. And then we're going to do it again six months later.”
SDKs are likely better than MCP sampling for remote server cases
“In remote cases, it's probably always better for you to bring an SDK, because you have full control, you can deploy it, you can deploy an API case, and maybe even charge someone.”
Properly implemented MCP sampling will become the definitive agent-to-agent protocol
“Like my, in, in some ways, you know, when you get it right this will be the real agent to agent protocol.”
The vast majority of current MCP usage is for context, not actions
“Most people use it for context. I think that's the vast majority of usage.”
Next MCP update will replace polling with server event callbacks and webhooks
“And then you can make an optimization, which we're going to do in the next round, which we didn't get around is like, okay, instead of having to pull every minute or hour or whatever you interval you choose, the server can call us events. Call you like, feel l…”
MCP strictly handles obtaining context, leaving selection and retention to applications
“And I think MCP I always see is like back to like, it's an application layer protocol. That's just how you obtain the context, how you select the context that the problem for the application. And that's the problem all the agent applications will have at the e…”
Passing raw OpenAPI specifications directly to LLMs bloats context and causes confusion
“If you put that into a model, like you will have a lot of bloat there too, right? Actually way worse. And funny enough, when people try to like map one to one things, often the model gets slightly confused because You have like search by name, search by ID, se…”
Anthropic and OpenAI are collaborating to build a unified AI UI standard
“And now one thing we just announced three weeks ago on the MCP blog is that we're actually working with all, all two of them together to build like a common standard.”
Regulated MCP clients may require formal certification and signatures within a year
“You might end up in a world really a year from now. There might be clients that have like certifications that they are in Amplik and get like a signature that they are like financial services, MCP clients, and they can prove it for the server. And only then th…”
Many current community-built MCP servers are mediocre
“I think we see a lot of mediocre MCP services. I mean, some very, very good ones.”
Anthropic remains fully committed to MCP following its foundation donation
“Like the commitment of Anthropic is the same, right? I'm still, We still have the same people I'm helping with the SDKs. We're still super committed in our products to MCP. I'm still the lead core maintainer. Nothing has actually changed.”
Anthropic always planned to put MCP into a neutral open-source foundation
“The first one is that on the MCP side, we always knew that we wanted to find a neutral home for MCP to make sure that the that the industry understands that this is stays open, that this is something safe to adopt.”
AAIF began when Block asked Anthropic about donating the Goose agent
“We got approached by our friends at block to discuss because they were looking into like donating goose, I think at the time. And so there was a question around doing something together. And then we approached open AI and they were very, very welcoming and lik…”
OpenAI is contributing AgentsMD to the new Agentic AI Foundation ecosystem
“In a similar way, the agentic foundation is, well, it's a foundation, but also it's like the starting point where I really look forward to other contributions, like starting with Goose, our own AgentsMD, where we're really open for like a lot of technical cont…”
AAIF launch saw more 24-hour inbound interest than any Linux Foundation project
“And 50 companies on day one, you know, a huge set of folks in line to participate and join my email inbox. I'm sure your guys are too, is like full in 24 hours. How do I participate? We want to contribute. How do I get in there? I've never seen that kind of in…”
Block's Goose was the first open-source agent to integrate MCP
“Goose was the first open source agent interface or agent that reached out to us and worked with us to integrate MCP. And I think Rad is actually like technically the first non-anthropic contributor to MCP ever on like day two or something like that, like very,…”
Block abandoned Goose's proprietary plugin ecosystem in favor of adopting MCP
“We had a version of Goose that was still, you can go check the GitHub history. It was there a little bit before MCP came out, and we were sitting there with like a plugin ecosystem who were like, this is awful. Like what, like, why would you, why would anyone?…”
Software standards and open source development must now be co-developed together
“I think the world of standards and open source development are just merging, right? You sort of co-develop these things together.”
Agent ecosystems need common standards but diverse, competing concrete agent implementations
“Like, agents.md is an example, which is open up any GitHub repository, it has this file, it works the same way. Like, if everyone sort of did their own thing there, That's very low value, potentially damaging in a way. But so there's commonality value, whereas…”
AAIF lacks a fixed project pipeline, selecting projects based on architectural principles
“I think we don't have a roadmap in the sense of like projects lined up, but I think what we have is principles by which we will select products to some degree.”
Vendor committee roadmaps typically fail in open-source foundations without technical curation
“It's more of a, not just more, but essentially a technical effort, not something where a committee of, you know, folks from vendors get together and say, well, my product should be in this roadmap, and that guy's product should be in this roadmap. It tends to …”
Technical participation in AAIF does not require formal foundation membership
“You don't need to become a member to participate on the technical side.”
Linux Foundation avoided frontier LLMs due to multibillion-dollar compute and data costs
“The Linux Foundation kind of took a pass on a, like, frontier model world, because in the open source space, having connection to an internet, the internet, And some intelligence and a computer. That's sort of entry. In the world of frontier LLMs, it's a compu…”
Agentic AI warrants a dedicated foundation because agent developers are a distinct discipline
“Agents are a distinct enough set of technology that it merits its own community.
[5135] Swyx (Shawn Wang): Separate from data and AI.
[5136] Jim Zemlin: Yeah, because like a PyTorch dev isn't really doing a ton of stuff in agent land, right?”
Linux Foundation separates technical governance from corporate funding to prevent pay-to-play
“And remember, we separate technical governance from
The governance of actually how money gets spent, because we don't want this sort of pay to play aspect of technology that tends to screw everything up.”
Zemlin: MCP and AAIF adoption will move ten times faster than Kubernetes
“I think there's a somewhat similar situation here with the caveat of saying, like, 10 times faster. Like, just day one, so much momentum around MCP. So much interest in this. And then also, 10 years of CNCF to sort of teach the developer community and the vend…”