The #telecom industry needs to engage rapidly with a new #platform technology called #MCP, which may form a central plank of future #AgenticAI architectures and #APIs. But it should not over-react or over-engineer. I'll do deep-dives later as I learn more, but MCP (Model Context Protocol) is moving so fast it needs immediate attention. It was announced by Anthropic in Nov'24 and has snowballed into a major "thing" In short, MCP servers give AI agents and #LLMs a well-structured description of, and access to, a set of available "resources". This could include APIs, tools, data sources or other AI functions. A common description is that MCP is an abstraction layer or "universal connector" that reduces the need for developers or IT asset owners to manually set up connectors for each of them. It could also be seen as a directory or registry of AI-accessible resources, or even a marketplace or appstore as an analogy. There will be many MCP servers, both in the cloud and in the enterprise's own systems. Some people describe it as being analogous to REST APIs, but designed for agents, with richer descriptions (context). There are also various attempts to create parallel agent-to-agent standards, such as Google's A2A. In telecoms, I see a range of possible uses for MCP: - Internal AI development by telco / vendor AI teams - Agentic discovery & consumption of telco and NetworkAPIs. Critically this could include both standard APIs (eg TM Forum or #CAMARA) and customer / specific APIs, and span telecom and enterprise networks, plus #CPaaS - Interaction with non-telco APIs (eg for specific verticals, apps, or adjacent sectors such as utilities and government) - Enablement of the #B2AI concept I introduced recently & discussed at the July #UnthinkableLab workshop with Andrew Collinson & Charlotte Patrick - MCP agents & servers inside network elements, such as RAN nodes, #5G / #6G core, home gateways, enterprise #WiFi or BNGs. That said, more deterministic (non-LLM / #GenAI) approaches may be more appropriate for low-latency or high-regulatory aspects such as #spectrum management. - *Maybe* a long-term post‑REST paradigm that could create a new AI front-end for systems, if we posit an "agent-native" world. There is already telecom work on MCP, including from TELUS, Telefónica, Vodafone, IETF, Sinch and The Linux Foundation #CAMARAProject Overall: - MCP is not (yet) a silver bullet. It’s a useful abstraction layer for APIs & future "Agentic-native" systems - Telcos should treat MCP as an urgent experiment for now - MCP and B2AI may become mutual catalysts - Security and regulatory concerns are *very* messy here. Beware - We're going to see lots of hybrids of REST, gRPC, MCP etc for different “users”: humans, apps, microservices, AI agents etc - No, telecom won't shape MCP. It's already too late - Laugh at anyone saying "Telco MCP market is worth $xxx billion" More to come from me. It feels important. Get in touch for private advisory & consulting.
Yep we’re building one, this is so obvious to anyone paying attention. The offer of discussing our strategy is still open…
Dean Bubley would you like to run a session at CASA25 about this?
I would calm the expectations. MCP is an interesting horizontal, there are others and there will be many more. I think your core message is that telecom should be aware of all of them, underneath the slogan level. MCP is not the answer, it could be a useful tool, make sure you understand its maturity and non-functional maturity (e.g. security). MCP is APIs for AI, if you are struggling with APIs then find somebody who isn't to help you...
Dear Dean, thanks for the post. We are building an open source project called OpenSlice under the umbrella of ETSI ( https://osl.etsi.org/ ) and we found MCP really valuable for the telco industry, thus we managed to bridge our TMFORUM models with MCP and expose them to LLMs and the results are really promising. if you have time Check our demo video here of how users can interact with the service catalogues via MCP with Claude desktop: https://www.youtube.com/watch?v=fSDDQvdbuME Source code and everything are open and available
Now we just need the big Telecom vendors to build MCP servers. As was mentioned by Geoff Hollingworth. MCP is a tool and not THE tool, which is probably why those vendors will not build MCP servers yet (or ever). Those system typically hold very sensitive data so Security is paramount. Things move rapidly and in many direction but with time things will settle and it will be clearer what frameworks are the most adapted for highly secure environments.
Big potential with Model Context Protocol (MCP). Its scalability and non-static benefits promise adaptability and growth. Crucially, though, MCP's substantial security implications demand proactive attention for responsible deployment. Prioritizing robust security is key as we embrace its power.
Hang on… this name is already taken: “The Master Control Program (MCP) is a fictional computer program and the primary antagonist in the Disney movie Tron.”
A2A complements MCP and there operators building control planes using agents. The technology moves so fast that within months (or weeks) solutions can be made. IoT and plug fests could be confined to the telco museum
Dean Bubley Good post. MCP is quickly making its way, but it alone won't be the holy grail. This is how I see. REST APIs ==> Application to application MCP ==> Agent to External tools and data A2A ==> Agent to Agent collaboration framework ACP ==> Standard interoperable agent interaction Telcos and 3GPP will start adopting these eventually, just like they started adopting SOA/REST APIs in Release 15. With that said, MCP has the potential to become superior alternative to REST APIs where applications dont need to re-code each time API versions change.
Innovation and Business Change | AI | IoT | 6G ecosystems | INSEAD
2moThanks Dean Bubley. My 2c. https://www.more-with-mobile.com/2025/06/ai-apis-and-iot.html