Architecture patterns and design decisions
Voir tous les tags
The Model Context Protocol (MCP) is a JSON-RPC 2.0 based protocol that standardizes how AI agents discover, authenticate with, and invoke external tools. It defines four phases — initialization, discovery, invocation, and streaming — over pluggable transports including SSE, WebSocket, and stdio. This article covers the protocol internals that matter for production deployments.
Three approaches dominate how AI agents call external tools in 2026: the Model Context Protocol (MCP), OpenAI Function Calling, and LangChain Tools. MCP is an open protocol for runtime tool discovery across any AI provider. OpenAI Function Calling is a proprietary API feature tightly integrated with OpenAI models. LangChain Tools is a framework abstraction that wraps tool definitions for orchestration pipelines. They solve different problems, operate at different layers, and can coexist in the same architecture.
Migrating from IBM DataPower or TIBCO requires separating gateway routing from protocol-specific functions. This guide covers a sidecar approach: deploy STOA for REST/JSON traffic, federate identity via OIDC, and keep legacy systems for B2B protocols where they excel.
IBM DataPower and TIBCO BusinessWorks represent two of the most deeply embedded integration platforms in enterprise IT. Both handle critical workloads — security token services, multi-protocol mediation, B2B gateway functions — that organizations depend on daily.
This guide provides a practical assessment of migration approaches for organizations evaluating modernization paths from these platforms.
MuleSoft migrations work best when you separate the API gateway layer from integration flows. Move Experience APIs (routing, auth, rate limiting) to open-source gateways while keeping Anypoint for DataWeave transformations and complex connectors.
MuleSoft Anypoint has become one of the most widely deployed integration platforms in enterprise IT. Since Salesforce's acquisition in 2018, the platform has deepened its ties to the Salesforce ecosystem while organizations face evolving requirements around AI agent support, European data sovereignty, and infrastructure cost management.
This guide provides a practical assessment of when MuleSoft migration makes sense, what the challenges are, and how to approach it without disrupting existing integrations.
Building a multi-tenant API gateway is one of the hardest infrastructure challenges in platform engineering. You need strong isolation between tenants, shared infrastructure for efficiency, and the ability to scale without multiplying operational complexity. After years of building multi-tenant API platforms — and applying those lessons to STOA on Kubernetes — here is what we learned.
This article is part of the Open-Source API Gateway 2026 series, covering architectural patterns for modern API infrastructure.
The European regulatory landscape has shifted dramatically for organizations managing digital infrastructure. DORA NIS2 compliance is no longer a future concern — it is an immediate operational requirement for any organization operating API gateways in financial services, healthcare, energy, or critical infrastructure sectors across the EU.
Let us say what many enterprise architects are thinking but few vendors will admit: the ESB is dead. The enterprise service bus — that monolithic integration middleware that defined the SOA era — has been in decline for a decade. What killed it was not a single technology but a series of architectural shifts: microservices, API gateways, event-driven architectures, and now the Model Context Protocol (MCP). Each shift made the ESB less relevant. MCP may be the final blow.
As AI agents move from demos to production, enterprises face a critical question: how do you give an LLM secure, governed access to your internal tools and data? The answer is an MCP gateway — a new category of infrastructure that sits between AI agents and the services they consume, enforcing security, observability, and policy at every interaction.