Awesome MCP › Agent Platforms

humanlayer/agentcontrolplane

⭐ 478 Go added to this list on 2025-05-14 repository created 2024-10-17

Agent Control Plane (ACP) is a cloud-native distributed agent scheduler designed for orchestrating AI agents on Kubernetes. It focuses on simplicity, clarity, and control, specifically targeting outer-loop agents that operate without direct supervision. These agents can perform asynchronous tool calls, such as requesting human feedback or delegating tasks to other agents, making ACP suitable for complex workflows involving human-in-the-loop and multi-agent coordination. ACP fully supports the Model Context Protocol (MCP), enabling seamless integration of various tools, humans, and other agents as part of the agent's operational environment. The architecture of ACP revolves around core objects including LLMs (Large Language Models), Agents, Tools, Tasks, and ToolCalls. LLMs represent the language model providers along with their API keys and parameters. Agents combine LLMs with system prompts and tools to perform tasks. Tools can be MCP servers, humans, or other agents, facilitating flexible and extensible interactions. Tasks represent user requests processed by agents, and ToolCalls are individual tool invocations during task execution. ACP is designed to run on Kubernetes clusters, supporting local setups via kind and Docker for container management. It requires an OpenAI API key or other compatible LLM providers for language model functionality. The deployment process involves applying Kubernetes manifests to create custom resource definitions (CRDs) and deploy the ACP operator. Users can define LLM resources, create agents referencing these LLMs, and run tasks that interact with the agents. The project embraces principles from the 12-factor app methodology adapted for agents, ensuring durability, reliability, and maintainability. It supports asynchronous execution of long-running tool calls and LLM inference, making it suitable for real-world AI agent orchestration scenarios. ACP is currently in alpha, with active development and community engagement through Discord and documentation resources. Keywords: Agent Control Plane, ACP, distributed agent scheduler, Kubernetes, AI agents, outer-loop agents, asynchronous tool calls, human feedback, Model Context Protocol, MCP, LLM, large language models, tools, tasks, tool calls, cloud-native, orchestrator, 12-factor agents, OpenAI API, human-in-the-loop, multi-agent coordination, Kubernetes operator, container orchestration, long-running tool calls, reliability, durability, alpha release.

https://github.com/humanlayer/agentcontrolplane

12-factor-agentsacpagent-control-planeagentsai-agentsalpha-releaseasynchronous-tool-callscloud-nativecontainer-orchestrationdistributed-agent-schedulerdurabilityframeworkhuman-feedbackhuman-in-the-loopkuberneteskubernetes-operatorlarge-language-modelsllmllmslong-running-tool-callsmcpmodel-context-protocolmulti-agent-coordinationopenai-apiorchestrationorchestratorouter-loop-agentsouter-loopsreliabilitytaskstool-callstools