AI & MCP Gateways

Gateways that route and govern LLM calls and agent access to MCP tools.

An AI gateway sits between applications and model providers to centralise keys, quotas, cost tracking, caching, and guardrails; an MCP gateway does the same for the tools agents call. Compare them on deployment ownership and license terms.

In dataset
7
Filter tools by type
Filter tools by cloud support
7 matching tools
More filters

Major vendor tools

Tools from the major platform vendors are listed first; independent and open-source options follow below.

Vendor

Edge-hosted AI gateway adding analytics, logging, caching, rate limiting, model fallback, DLP and guardrails in front of LLM providers.

Proprietary
SaaSSOC 2License medium

Filtered open source and third-party tools

5 matching non-vendor tools in this filtered view.

agentgateway

Linux Foundation

Open Source

Rust proxy for agent traffic that governs MCP, A2A and LLM calls with CEL-based RBAC, budgets and guardrails; runs standalone or on Kubernetes.

Apache 2.0
Self-hostedLicense low
Open Source

Open-source docker mcp CLI plugin that proxies MCP servers run in isolated containers, with central secrets, OAuth and call tracing.

MIT
Self-hostedLicense low
Commercial

AI and MCP governance layer on Kong Gateway: multi-LLM routing, semantic caching, prompt guards, PII redaction and MCP OAuth.

Apache 2.0 core + Enterprise plugins
SaaSSelf-hostedHybridLicense medium

LiteLLM

BerriAI

Open Source

Self-hostable AI gateway and Python SDK that calls 100+ LLM APIs in OpenAI format with virtual keys, spend tracking and guardrails.

MIT core + EE paths
Self-hostedSOC 2License medium

Portkey AI Gateway

Portkey (Palo Alto Networks)

Open Source

Open-source AI gateway that routes to 250+ LLMs with fallbacks, caching and 50+ guardrails; managed SaaS and private-cloud options.

MIT
Self-hostedSaaSHybridSOC 2License low

Important notes

Warning

Azure API Management AI gateway: The AI gateway is part of API Management, not a separate offering, and availability varies by service tier. The unified model API and the AI gateway in Microsoft Foundry integration are in preview (as of 2026-09-30).

Warning

Cloudflare AI Gateway: Gateways first created on or after 2026-09-24 use Workers Logs pricing and retention; older customers stay on Legacy Logs.

Warning

Docker MCP Gateway: The OSS CLI plugin is generally available; the MCP Gateway as part of Docker AI Governance is invite-only (as of 2026-09-30).

Warning

Kong AI Gateway: Open core: Kong Gateway and six basic AI plugins (ai-proxy, ai-prompt-guard, ai-prompt-decorator, ai-prompt-template, ai-request/response-transformer) are Apache 2.0, but advanced plugins such as AI Proxy Advanced and the MCP gateway features need the paid AI Gateway Enterprise offering.

Warning

LiteLLM: Open core: code outside the enterprise/ directory is MIT, but enterprise/ is under the BerriAI Enterprise License, which allows production use only with a paid BerriAI subscription (dev/test use is free). Enterprise features such as SSO and audit logs sit behind that license.

Warning

Portkey AI Gateway: Palo Alto Networks completed its acquisition of Portkey on 2026-05-29, and Portkey is now the AI gateway for Prisma AIRS. Gateway 2.0, which merges Portkey's production gateway into the open-source codebase, has been a pre-release on the 2.0.0 branch since 2026-03-24.
2026-08-27
agentgateway
agentgateway

agentgateway 1.5.0 added API-key-scoped LLM budgets, native Gemini APIs, and guardrails on tool calls, with breaking JWT and token-count changes.

Source

Frequently asked questions

  • What does an AI gateway do?

    It is a proxy between applications and model providers that centralises API keys, routing and fallback across models, rate limits and budgets, caching, logging, and guardrail checks, so every team does not reimplement them. Many now also expose or govern MCP tool traffic.

  • What is an MCP gateway?

    A gateway for the Model Context Protocol: agents connect to one endpoint, and the gateway decides which MCP servers and tools each agent may use, injects credentials, and logs every tool call. It is the control point for agent tool access in the same way an API gateway is for APIs.

  • Build on an API gateway or adopt a dedicated one?

    If an API management platform is already the enterprise standard, its AI features keep one policy surface. Dedicated AI gateways usually move faster on model and provider support. Check licensing carefully: several open-source gateways keep enterprise features under a commercial license.

Explore adjacent hubs

Compare the active category against the platform foundation layer, the cross-category updates feed, and the sourcing/contribution guide.