node-wire + ToolHive provide the interoperability layer that enterprises need to adopt MCP at scale
Enterprise AI has a quiet dependency problem. While the models get attention, the real value shows up when an agent can access the systems where the work actually happens. This could be an EHR, a core banking system, an identity provider, or a case management system. Consistently and securely reaching those systems is a challenge on its own, and this is where many AI projects get stuck – they lack a connection layer. That reusable connector layer is what AOT Technologies builds.
node-wire is an open-source connector framework published by AOT Technologies under an Apache 2.0 license. It is built in Python for engineers who are developing AI systems, enterprise integrations, and reusable connectors. It acts as a connector execution framework rather than a workflow manager. node-wire provides a reliable, observable, and secure layer that sits between callers, whether those are human applications, backend services, or AI agents, and external SaaS and enterprise systems. As a general-purpose framework, it complements MCP platforms.
The last-mile integration gap
ToolHive, created by Stacklok, is an open-source MCP platform that allows for the secure deployment, management, and governance of MCP servers and agent skills across local, desktop, and Kubernetes environments. In this system, MCP servers serve as the standard interface for AI agents to find and use enterprise tools.
However, not all enterprise integrations can readily function as MCP servers. Many external systems use traditional REST APIs, gRPC services, databases, message queues, proprietary SDKs, or legacy interfaces. To meet these integration needs, a secure, open-source connector layer is required.
node-wire can provide this connector and interoperability layer for ToolHive. It supports secure communication with enterprise systems through various protocols and authentication methods. This allows agent-to-tool connections managed by ToolHive to work with systems that do not yet provide native MCP interfaces. node-wire includes a documented reference architecture: each connector comes as a per-connector MCP server container that ToolHive can register, secure using its own secret store, and run. This setup enables agents to connect to enterprise systems through governed MCP tools. It also provides a reference agent that demonstrates the process from start to finish.
To support workflows in healthcare, node-wire offers FHIR connectors for EHR systems such as Epic and Cerner, with particular attention to essential capabilities, including retrieving the patient record. In fact, the connector implementation is intentionally restricted to support specific scenarios rather than covering all capabilities described in the FHIR standard. The reference scenario involves the collaboration of several node-wire connectors that are exposed as MCP tools. Once these tools are available, the agent knows which tools are required and dynamically connects them to accomplish the task. Given a natural-language command, the agent retrieves a patient record via the FHIR connector, uploads it via the Google Drive connector, and finally sends a notification via the SMTP connector. For end users, it eliminates the need for manual, multi-step data handling and context switching, turning complicated cross-system workflows into safe, immediate operations.
The MCP Builder Integration
The MCP Builder is the emerging link between the two. It is a stated direction rather than a delivered capability today, with full integration planned for early August 2026. mcp-builder is Stacklok’s open-source pipeline that simplifies a large OpenAPI specification into a small, well-defined tool set. mcp-builder is used by AOT Technologies not for running, but as a scaffolding and curation engine, while the transformer, which generates native node-wire connector packages, is being developed. Curation happens in mcp-builder while execution occurs in the node-wire runtime. Every generated connector follows the same execution rules, error handling, authentication, and telemetry as a hand-written one, and is accessible via REST, gRPC, and MCP like any other. The goal is to streamline the process from API specification publication to implementation of the governed connector, as well as increase interoperability within the wider MCP ecosystem.
What node-wire brings
node-wire is organized into three distinct layers. A Runtime manages everything that should act the same across integrations. Connectors handle only what is truly system-specific: the request and response schemas and the calls to a vendor API or SDK. Bindings manage protocol exposure. Since the same connector definition is projected into REST routes, gRPC methods, and MCP tools at the same time, an Organization writes an integration once and uses it in a web application, a backend service, and an LLM agent, with no differences in validation, error handling, or telemetry.
Eight connectors come with the framework: HTTP Generic, SMTP, and SMART on FHIR connectors for Epic and Cerner. Together, they cover common enterprise categories, including payments, CRM, document storage, messaging, and healthcare EHR access, with a documented model for creating more.
node-wire also maintains a privacy-by-design approach. This solution does not conduct any telemetry or phone-home actions, and because the framework does not include a built-in database, it does not store transactional data unless the deployment organization adds its own storage layer. Moreover, there is a platform-wide sanitizing filter that deletes all PHI-like fields from the logs by scrubbing specific field names and values at the telemetry boundary. Together with the FHIR connectors and the HIPAA considerations documentation, it makes the framework suitable for use in regulated environments. The authors of the project admit that these elements alone cannot guarantee HIPAA compliance, as it depends on the whole system, but they remove the framework from this list of barriers.
A financial-services example
To illustrate how this works, consider a financial services organization that would need to establish connections among its AI agents, SMTP Servers, graph databases, and EHR platforms. Instead of developing custom connections that will lead to disjointed error management and security gaps, this organization can integrate ToolHive and node-wire to establish a standard infrastructure. In this infrastructure, ToolHive will work as the governance control plane for MCP tools, while the node-wire connectors will handle the actual implementation.
“Agent-to-tool connections can create outsized value or immeasurable harm. Enterprises are under pressure to ensure every agent action is secured and approved. And not only do they need to exert control over their data, the rate of change in AI demands that enterprises remain in control of their destiny. All that points to open source solutions like node-wire and ToolHive that multiply the value of AI agent investments and ensure enterprises remain platform agnostic.”
~ Craig McLuckie, founder and CEO, Stacklok
A shared bet on how enterprise AI gets built
As AI enters production, organizations are investing in both models and the infrastructure to run them responsibly. This includes secure runtimes, governance, observability, and reliable connectivity. node-wire focuses on the connectivity layer, which is often ignored and can be expensive if done wrong. It provides this security through tangible runtime capabilities: transport security using JWTs and API keys; scope-based authorization via tools; and outbound integration security using static tokens, OAuth 2.1 with PKCE, and cloud secret managers (AWS, Azure, GCP, Vault).
“The bottleneck in enterprise AI is proper integration. Reusable connector frameworks are shifting away from plumbing that connects everything, to strategic infrastructure with governance and trust built in. This foundation provides users with the confidence their organizations have the best tools in place, for holistic AI integration.”
~ Praveen Ramachandran, Co-Founder and Director at AOT Technologies
Looking ahead
MCP is developing rapidly. AOT Technologies expects reusable connector frameworks to become a standard part of enterprise AI infrastructure, rather than something each team has to create from scratch. AOT Technologies is building node-wire openly, and we are happy to collaborate with partners like Stacklok as compatibility among node-wire, the MCP Builder, and ToolHive continues to improve.
Whether you want to test a pre-built connector in a sandbox, discuss an integration use case for your stack, or contribute to our open-source framework, we welcome you to connect with our engineering team to get started.
Author bio:
Peter Famiglietti leads global partnerships and alliances at AOT Technologies, where he drives go-to-market strategy for channel expansion of AOT Technologies’ services and product portfolio. Peter holds a BSc in Engineering from McMaster University and an MBA from the University of Western Ontario.
Vinaayakh Gopalakrishnan is a Software Architect at AOT Technologies with nearly a decade of engineering experience. A core contributor to node-wire and m8flow, he works on system design and integration architecture and on making complex systems simple to connect.