Agent & MCP Tooling

LangChain is an open-source Python framework that provides composable building blocks for chaining LLM calls, tools, retrievers, and memory into production AI applications. It includes LangGraph for stateful multi-agent orchestration and 1000+ integrations with model providers and data stores.

Open-source MCP gateway and Docker CLI plugin for running, grouping, and exposing MCP servers through Docker-managed profiles and catalogs. It powers Docker Desktop's MCP Toolkit and also runs independently against a Docker engine.

AGPL-3.0 hybrid platform combining LLM gateway, MCP orchestrator, and chat UI into one Kubernetes-native stack. Orchestrator spawns each MCP server as an isolated K8s pod — strong isolation, but K8s-only in practice.

Open-source gateway and registry for MCP servers, A2A agents, and REST or gRPC APIs. It centralizes discovery, auth, routing, and observability for agent tool access across local, container, and Kubernetes deployments.

Open-source MCP gateway for routing and governing Model Context Protocol servers. It proxies configured MCP servers, applies security plugins to requests and responses, and can scan server reputation before loading them.

MIT AI gateway from Portkey AI that routes traffic across 1,600+ LLM endpoints through one OpenAI-compatible API. Combines routing, virtual keys, guardrails, caching, observability, and MCP gateway support in a Node.js / Go proxy that runs on Docker or Kubernetes.

LLM gateway from Maxim AI written in Go, with OpenAI-compatible routing, virtual keys, budgets, MCP client and server, and a plugin-based governance pipeline. Apache 2.0 core with a commercial Bifrost Enterprise tier.

Apache 2.0 AI-native API gateway from Alibaba, built on Envoy, with multi-provider LLM routing, built-in MCP hosting, OpenAPI-to-MCP conversion, and token rate limiting. Runs as a single Docker container or on Kubernetes via Helm.

Open-source MCP aggregator from metatool-ai that federates upstream MCP servers into namespaced endpoints for MCP clients. Three-level hierarchy of Servers, Namespaces, and Endpoints, with per-endpoint tool ACLs and generic OIDC discovery for admin login.

Lightweight MPL-2.0 MCP gateway that aggregates upstream servers under one API-key-authenticated endpoint. YAML config file, no database required for basic deployments, single Docker container.

Apache 2.0 Rust agent gateway covering MCP, A2A, and LLM protocols, governed by the Linux Foundation with AWS, Cisco, IBM, Microsoft, Red Hat, and Solo.io participating. Kubernetes Gateway API-first with a standalone binary mode stabilizing through pre-1.0.

Self-hostable LLMOps platform from LangGenius with a visual agent builder, RAG pipeline, multi-provider gateway, and prompt management. Uses the Dify Open Source License — modified Apache 2.0 with a non-compete clause prohibiting multi-tenant SaaS resale.

Self-hostable web UI for LLMs from the Open WebUI project. Multi-user chat with generic OIDC authentication, shared workspaces, document RAG, MCP client support, and Python-based Pipelines and Functions for custom extensions.

Open-source AI chat platform from Danny Avila that mirrors the ChatGPT interface with multi-provider routing, plugins, MCP client, and OpenID Connect login. Docker Compose reference deployment uses MongoDB with optional MeiliSearch for conversation search.

Python-based AI gateway from BerriAI that proxies 100+ LLM providers behind one OpenAI-compatible API with virtual keys, hard-enforced budgets, cost dashboards, and an MCP federation endpoint. Core is MIT; SSO, JWT-to-key mapping, and audit retention sit in LiteLLM Enterprise.