agents MCP Client: Multi-Harness Plugin Marketplace for LLMs
agents: A Unified Agentic Plugin Marketplace for LLM Harnesses
The agents MCP Client delivers a multi-harness agentic plugin marketplace, offering a comprehensive suite of 83 plugins, 191 agents, 155 skills, and 102 commands. This client is designed for developers who need production-ready agentic workflow building blocks, consumable natively across various LLM environments including Claude Code, OpenAI Codex CLI, Cursor, OpenCode, and Gemini CLI.
MCP Integration via Plugin Marketplace
agents integrates with MCP primarily through its plugin marketplace, making its extensive collection of agentic components directly accessible within supported LLM harnesses. For Claude Code users, integrating agents is a straightforward two-command process. You first add the wshobson/agents marketplace, then install specific plugins like python-development or any of the other 82 available options.
/plugin marketplace add wshobson/agents
/plugin install python-developmentThis approach allows Claude Code to leverage the full capabilities of agents's agentic workflow building blocks directly within its environment.
Multi-Harness Native Support
A core strength of agents is its "one source-of-truth, five harnesses" architecture. All agentic components — plugins, agents, skills, and commands — are defined from a single Markdown source located in the plugins/ directory. From this unified source, agents generates idiomatic, harness-native artifacts for each supported platform, ensuring optimal performance and integration rather than a lowest-common-denominator translation.
Beyond Claude Code, agents supports OpenAI Codex CLI, Cursor, OpenCode, and Gemini CLI. For these harnesses, the setup involves cloning the repository and running a make generate command, specifying the target harness:
gh repo clone wshobson/agents ~/agents
cd ~/agents
make generate HARNESS=<codex|cursor|opencode|gemini>This process ensures that each harness receives artifacts tailored to its specific requirements, maximizing the utility of the agentic components within its native context.