The Linux Foundation's CAMARA project recently published a white paper outlining how telecommunications network capabilities can be exposed to Artificial Intelligence (AI) systems through the Model Context Protocol (MCP), addressing a fundamental discoverability challenge as AI tools increasingly shape how software gets built and integrated.
The 11-page technical position paper, In Concert: Bridging AI Systems & Network Infrastructure through MCP, describes an architecture where network operators expose capabilities – device location verification, bandwidth optimization, fraud detection signals – as tools that AI systems can discover and invoke programmatically. The integration leverages MCP, a protocol originally developed by Anthropic and now housed under the Linux Foundation's Agentic AI Foundation, which provides standardized interfaces between AI models and external systems.
Ranny Haiby, CTO of Networking, Edge & Access from the Linux Foundation explains:
If you don't have MCP presence, you're invisible. It's not anymore people writing code, it's AI tools writing AI agents. If you don't have MCP presence, you simply are not discoverable.
That reflects how development workflows are changing. As AI-assisted development becomes more common, the interface layer that AI tools can see starts to influence what gets used, integrated, and operationalized. For telecommunications operators, MCP provides a route to ensuring network capabilities remain accessible inside modern development workflows.
The standardization challenge
CAMARA launched in 2022 to address API fragmentation across telecommunications operators. Each network provider historically exposed capabilities through proprietary interfaces, forcing developers to write custom integrations for Deutsche Telekom, then rebuild for Verizon, then again for Vodafone. The project aims to define operator-agnostic APIs that work consistently across carriers – a "write once, run anywhere" approach for network services.
MCP adds urgency to that standardization effort. The protocol has seen rapid industry adoption since Anthropic's initial release in November 2024, with the Linux Foundation citing more than 10,000 published MCP servers. Major AI vendors including OpenAI and Microsoft Azure have integrated MCP support, and the protocol joined the Linux Foundation's Agentic AI Foundation in December 2025 alongside OpenAI's AGENTS.md specification and Block's Goose framework.
The pressure is practical as well as competitive. Herbert Damker, CAMARA Technical Steering Committee Chair and Lead Architect for Infrastructure Cloud at Deutsche Telekom, describes the global scale dynamics:
Hyperscalers have globally the same API for their platforms. They are proprietary for their platforms, but they are large enough that they can offer a global service alone. Telco providers have the problem that there are 700 in the world. The largest 20 are covering already a lot of the users, but they have to offer the same service if they want to be in a global scale play.
The white paper positions an MCP server as a translation layer between AI applications and CAMARA network APIs. AI systems equipped with MCP clients can discover available network capabilities, request specific services, and receive responses – all using standardized tool descriptions designed to behave consistently across different AI models and network operators.
Network capabilities for AI systems
The paper shares categories of network intelligence that applications typically cannot access from above the connectivity layer.
Identity verification and fraud signals – CAMARA's number verification API confirms that a mobile number belongs to a specific device through either network IP address matching when devices use mobile data, or cryptographic authentication via SIM card when connected to WiFi. The specification is designed to reduce reliance on one-time password authentication. Google is integrating this capability into Firebase, making it available to applications using that authentication platform.
SIM swap detection queries whether a SIM card was replaced within 48 hours, flagging suspicious high-value transactions. Damker notes that common fraud scenarios involve attackers obtaining replacement SIM cards for existing numbers to bypass authentication controls.
Quality on Demand (QoD) – rather than simply requesting increased bandwidth, applications can receive real-time signaling about impending congestion, enabling preemptive adjustments in user experience. Damker describes a gaming implementation:
"We're planning to offer together with the gaming provider not giving more bandwidth, but a slice which is signaling in real time on the packet markers when buffering would happen shortly. So the application can react before it will be queued."
This approach prioritizes latency stability over resolution for interactive applications. Deutsche Telekom also offers a production service for live video streaming that guarantees upstream bandwidth for journalists broadcasting from mobile devices – "more or less replacing the satellite van," according to Damker. German broadcaster RTL uses this service for field production.
Edge discovery – because network topology is opaque from outside the infrastructure, applications cannot easily determine optimal deployment locations based on actual connectivity paths. Damker explains:
"Networks are from outside not transparent. You can go by distance if you know the location of the device, then you can say, okay, the next one will be Frankfurt for a German user. But from a network perspective, that can be a completely different location like Amsterdam is better suited and nearer connected to the device."
CAMARA's Edge Discovery APIs expose this intelligence, helping applications identify optimal edge computing nodes based on network performance characteristics rather than geographic proximity.
Security posture and production readiness work
The CAMARA–MCP combination sketches how agentic systems might consume network capabilities in a way that is discoverable, policy-aware, and consistent across operators. It also expands the security and governance surface area. MCP servers act as the layer that advertises available tools and invocation parameters, placing greater emphasis on implementation practices alongside API design.
The white paper positions MCP support as an engineering program with explicit objectives – including security guidelines, standardized MCP tooling for CAMARA APIs, and quality requirements needed for production-grade implementations.
Authentication and consent are central to that agenda. CAMARA's OAuth 2.0/OpenID Connect security profile aligns conceptually with MCP's authorization approach, but implementation complexity remains. "Three-Legged Access Tokens" involve coordination between users, applications, and authorization servers. Because AI systems cannot participate directly in interactive OAuth flows, the MCP layer must handle that complexity while preserving privacy and consent requirements.
When asked about authentication flows, Damker notes:
"How to couple us really on the MCP, that is the work which we have to do."
Early implementation work focuses on tool exposure, discovery, and consistent API usage, while operational models continue to evolve as deployments mature.
Pricing, SLAs and operational expectations
The paper also highlights an evolving commercial model. Hyperscalers have conditioned developers to expect self-serve pricing, published per-call consumption models, and defined production SLAs with explicit availability commitments. Telecommunications APIs have traditionally been consumed through enterprise contracting and partner channels.
Even when network APIs are exposed through hyperscaler marketplaces such as AWS, Azure, and Google Cloud, they often remain closer to enterprise sales models than to cloud-style elasticity. The ecosystem is still defining cost models and SLA expectations, particularly as AI systems generate significantly higher transaction volumes than human users.
Tanja de Groot, Open APIs Chief Architect of Nokia, acknowledges the ecosystem work ahead:
There is, of course, business aspects like charging, and so how do you make people pay for these types of models? That's part of the ecosystem that will need to be worked out.
The authentication discussion points to a broader shift in design. OAuth and OpenID Connect patterns assume a human in the loop to consent and authenticate. MCP introduces non-human clients – agents and tools – that can discover and chain capabilities autonomously.
This raises implementation questions around token lifecycle management, scope constraints, and tenant boundaries when MCP servers front multiple operator APIs. Production deployments also require clear expectations around logging, observability, and shared responsibility across telecommunications operators, MCP implementers, and enterprise users.
CAMARA + MCP provides a view into how AI agents may integrate with network and identity infrastructure. Fraud engines, login flows, and workload placement could query network-verified signals in real time rather than relying on inference or third-party approximations.
Enterprise procurement models emphasize transparent pricing and defined SLAs, particularly where APIs form part of critical security or customer-experience workflows. Architecture teams must also establish guardrails governing agent access to network capabilities, including scope controls, rate limits, and observability requirements.
Industry precedent and standardization risks
CAMARA is not the first attempt to standardize operator APIs at global scale. Earlier initiatives – Parlay/Parlay X, GSMA OneAPI, and Wholesale Applications Community – struggled with complex specifications and uneven implementation. Developers often preferred aggregators offering more predictable consumption models.
The project aims to avoid those outcomes by anchoring in open source, aligning with GSMA Open Gateway, and integrating with the AI tooling ecosystem. Achieving operational consistency across operators and regions remains a focus of the effort.
Arpit Joshipura, General Manager of Networking, Edge and IoT at the Linux Foundation, describes success across technology, ecosystem, and standardization:
"There's a success on the technology parts, which is how the architecture evolves in terms of the agents, the MCP layer, the API layer, and the network below. The second criteria for success is the ecosystem... are we getting the right number and type of aggregators? Are we getting marketplaces inside hyperscalers? And then I think the other thing we are monitoring is making sure that we don't have a global fragmentation."
De Groot confirms that user-side take-up – growth in the number of developers and applications actually consuming the APIs — remains a key success metric alongside infrastructure availability.
My take
Network operators hold authoritative signals – in both senses of the word – from device identity and SIM state to network conditions and topology. Exposing those signals through operator-agnostic APIs, and making them discoverable through MCP, aligns telecommunications infrastructure with emerging AI-assisted development practices.
Number verification and SIM swap signals provide fraud teams with stronger inputs than many third-party alternatives. QoD moves beyond "more bandwidth" into experience-aware signaling that applications can act on. Edge discovery replaces guesswork about deployment location with network-informed placement decisions.
Even at the white paper stage, this is dense engineering territory. Bringing telecommunications infrastructure, identity assurance, and agentic AI into the same control plane introduces a level of architectural complexity that enterprises will need time to unpack.
The next phase is turning that promise into production-ready services through security guidance, standardized tool definitions, and consistent operational expectations. CAMARA is explicit that production readiness depends on authentication flows, consent handling, quality requirements, and conformance expectations that work consistently across operators. The key indicators to watch will be production-grade implementations, published pricing models, defined SLAs, and consistent behavior across operators.