Standardizing for Commoditization — Anthropic’s Model Context Protocol (MCP)

Originally posted on Medium: Standardizing for Commoditization — Anthropic’s Model Context Protocol (MCP) | by Sam Bobo | Dec, 2024 | Medium

Imagine a world where every plug fits every socket, where your smartphone seamlessly connects to any Wi-Fi network, and where products from different corners of the globe work together flawlessly. This isn’t a utopian dream — it’s the reality made possible by standards. These often-overlooked guidelines are the unsung heroes of our interconnected world, quietly ensuring that everything from our morning coffee machines to the global internet infrastructure operates smoothly and efficiently. During day-to-day life, we take standards for granted but never take a step back to appreciate the true impact of standardization.

Take for example, the Hypertext Transfer Protocol. Put another way, HTTP, those four characters (five if you include “S” for secure) followed by “://” and the infamous www for worldwide web that begin a website URL. HTTP was a protocol that set the foundation of data communication in the internet, from defining how messages are formatted and transmitted to how web servers should render information on a screen. We’ve also experienced standardization in the Universal Serial Bus, again, known as a USB chord which became the standard for cables and connectors for connection, communication, and supply. Most recently, USBs have become further standardized, particularly around the type C connector (USB-C).

Standardization drives innovation by forming a foundational ecosystem for innovation, quality, and consistency, yet, when viewed through a competitive lens, takes a interesting story worth exploring. Lets dive into three examples of standardization over time:

• The Past: the Media Resource Control Protocol (MRCP) pioneered by Nuance used historically for speech engines functioning on a telephony channel, making legacy interactive voice response systems over the past few decades
• The Present: USB-C — the type C USB connector with Apple at a major disadvantage
• The Future: Model Context Protocol (MCP) recently announced by Anthropic

Each of the three time periods tells a story about standardization and the influence establishing these standards has on the market. With Anthropics’s recent announcement about MCP, I thought exploring the concept of standardization would be pertinent to get a sense on why Anthropic made such an announcement, and it all comes down to competition. (commoditizing supplements and cementing advantages)

The Past — Media Resource Control Protocol (MRCP)

A fascinating example of standardization, especially in my line of work, is the Media Resource Control Protocol, or MRCP. First published as an RFC in 2006, The MRCP protocol started as a communication protocol used by speech servers to provide Conversational AI based services such as speech recognition (speech-to-text) and speech synthesis (text-to-speech). The standard was co-authored by Nuance Communications, Cisco Systems, and SpeechWorks (acquired by Nuance). Together with VoiceXML as the language describing a voice browser, a program that creates the logical structure for voice powered applications, and SIP (Session Information Protocol) to direct real-time communication sessions over IP networks created the modern infrastructure for Interactive Voice Response (IVR) systems.

Standardizing on the MRCP protocol facilitated a vast ecosystem of suppliers of both artificial intelligence capabilities as well as telephony. Enterprises could chose vendors such as Cisco, Avaya, and Genesys paired with Nuance or other AI vendors to create a comprehensive IVR solution that was interoperable and changeable at each component of the value chain. This further improved efficiency, drove down cost, and decreased the complexity of launching an IVR. Effectively speaking, the transportation of voice became commoditized due to standardization of MRCP.

Within this ecosystem thrived innovation as new entrants were able to join, either from the AI module or telecommunications (albeit significantly harder due to the level of infrastructure cost of towers, a major barrier to entry) but gave rise to these telecommunication companies as platforms to select any AI vendor as the best-preferred option. Over time, extensions were built upon MRCP such as magic word or custom parameters that created a level of lock-in. In parallel, IVR vendors started competing in the app level by building abstraction layers above VoiceXML in low(er)-code paradigms to introduce additional lock-in.

Fast forward, workloads started shifting to the cloud as compute became cheaper, capital expense shifted to operational expense, and the benefits of the cloud (cost, scalability, etc) became realized. At that juncture, Google made moves to introduce gRPC as a successor to MRCP and improved voice transmission as well as APIs into Conversational AI systems.

In summary — the MRCP standard aided in the flourishing of intelligence speech systems but ultimately usurped during technological paradigm shifts such as cloud computing that introduced new standardization protocols fitting that paradigm as well as competition vertically integrating across AI, Application, and Telephony.

The Present — USB-C

Let’s recall the first paragraph of this piece, where we imagine a world where every plug fits every socket. That was the mission of IBM, Microsoft, NEC and Nortel who helped pioneer the Universal Serial Bus, more commonly known by its acronym, USB. The objective of developing the USB was to create a universal standard to simplify connections between computers and peripherals. The first release of USB came in 1996 offering transfer rates of 1.5 Mbps, with USB 2.0 released in 2000 supporting 480 Mbps and the chord still in use today by most suppliers working with USB type A. In 2008, USB 3.0 was released and introduced transfer rates of 10 Gbps (massive increase!) and the USB type C, that oblong connector that is now used for modern electronics and quick-charge capabilities. With most electronics purchased today — from speakers to camping lanterns — anything with a charging cable uses the USB standard.

Apple, the outlier, sought to create its own power adapter, the lightning cable. The Lightning cable was first introduced by apple in 2012, after USB 3.0 was released, and was proprietary to apple. It included a reversible connector (more user friendly than USB A), compact in size for portable devices, durable, and “Made for iPhone/iPad/iPod” certified, so… Apple only.

What happened to Apple, however, was a European Union ruling in October 2022 that passed a law requiring all new mobile phones, tablets, and cameras sold within the EU block to run on USB-C connectors by the end of 2024. This basically enforced the standard set forth back in the early days of technology. This largely was targeted at Apple, but quoting the press release:

Parliament’s rapporteur Alex Agius Saliba (S&D, MT) said: “The common charger will finally become a reality in Europe. We have waited more than ten years for these rules, but we can finally leave the current plethora of chargers in the past. This future-proof law allows for the development of innovative charging solutions in the future, and it will benefit everyone — from frustrated consumers to our vulnerable environment. These are difficult times for politics, but we have shown that the EU has not run out of ideas or solutions to improve the lives of millions in Europe and inspire other parts of the world to follow suit”

Standardization, in this format, protected the environment and consumers who were required to effectively have a “bag of chords” disjointed and not interoperable. Apple was the largest hurt as it removed a proprietary capability they sought for their devices, however, I personally believe that standardization will be better in the long run but will remove Apple’s ability to innovate in charging chords. The question becomes, will standardizing on USB allow for greater capabilities such as data transfer speeds, quick power, and the like? My thought is yes.

The Future — Model Context Protocol (MCP)

So this brings us to the Future, centered around Artificial Intelligence and the rational behind this blog post. On November 25, 2024, Anthropic released a blog post announcing a new open standard initiative - The Model Context Protocol (MCP). Quoting from the introduction:

Today, we’re open-sourcing the Model Context Protocol (MCP), a new standard for connecting AI assistants to the systems where data lives, including content repositories, business tools, and development environments. Its aim is to help frontier models produce better, more relevant responses.

Understanding MCP requires some background into the entire AI ecosystem. Lets take some of the key players: OpenAI, which is backed by Microsoft and access to Bing as a search engine, Gemini powered by Google (self explanatory), Llama by Meta which is open source but heavily used in their case for social posts, and Anthropic, which, I presume, lacks the primary access to rich, up-to-date, information that a search engine or social network would provide. (Note: It goes without saying I am overlooking smaller players list Mistral in this analysis in an effort to use Anthropic as the key example within the cohort). As we’ve explored in the aforementioned examples, standardization provides a modality for an ecosystem to flourish and commoditizes part of the value chain in the process, which could take away advantages from others (e.g Apple).

Anthropic smartly recognized that LLMs (or the pursuit of AGI broadly) still requires grounding in some capability, especially as these engines are probability and statistic machines. Retrieval Augmented Generation (RAG) is the safest approach to contextual generation and fits nicely on the Roadmap to Trust for consumers:

Content summarization is a highly effective Generative AI use case namely due to the popularization and adoption of a technique called Retrieval Augmented Generation (RAG). RAG takes a corpus of knowledge — PDFs, websites, documents — indexes them, and vectorizes them within a vector database. With RAG, the knowledge corpus is limited or constrained to the corpus of knowledge pre-built, not with the vast training underpinning the LLM. Thus, content summarization can be deemed as stable.

The strategy I am alluding to is the commoditization of data. In order to understand the argument being made, one must make a broad assumption that organization and consumers alike will not utilize the Three Layers of Ground Truth for AI, but rather will simply bypass the framework and move directly to a foundational model plus RAG, which is a simplistic approach and one that Anthropic is betting on. Why commoditize data?

Lets examine some of the key players — Microsoft and Google. Microsoft provides Copilot as the Generative AI assistant included in all facets of their products, from Office to Dynamics to even Azure. Microsoft Office 365 includes OneDrive, and SharePoint as common data repositories for business related work as well as a massive portfolio of 1400+ connectors via the Power Platform and Copilot Studio. Google, similarly, has Google Workspace and Vertex AI alongside Gemini to help perform RAG-based searches on common business data stored in office files. Removing both a barrier to access for common information (lets also include the face that there is lock-in at the cloud provider level as egress fees (data transfer out) matter (as opposed to ingress or entry fees are non-existent). This strategy is blatantly obvious, again, from the blog post:

Claude 3.5 Sonnet is adept at quickly building MCP server implementations, making it easy for organizations and individuals to rapidly connect their most important datasets with a range of AI-powered tools. To help developers start exploring, we’re sharing pre-built MCP servers for popular enterprise systems like Google Drive, Slack, GitHub, Git, Postgres, and Puppeteer.

The challenge, however, I foresee is the creation of MCP Servers that are shown within the documentation, as I believe there will be high development costs and yet another framework to use. Anthropic will win should they be able to overcome the Aggregator that is OpenAI but I anticipate succeeding will take regulatory power to both shift users over to Claud and create demand that forces the hand of large tech players to comply.

If Anthropic succeeds, they will need a significant design advantage to overcome the ease of use that PowerPlatform and other connector-based DIY solutions have. Employing RAG would be beneficial to gain traction on Generative AI based solutions broadly, albeit with a standard use of primitive use cases at the moment before logical systems and agentic architectures arise, but hey, I’m an advocate of AI and it would please me to see adoption. I do have my doubts, as I mentioned, for Anthropic pulling this off but kudos for trying.