vLLM Agentic API

Rust gateway in front of vLLM that owns stateful agentic APIs (OpenAI Responses-style), including conversation state, tool-call loops, and MCP host allowlists. Moves multi-turn agent orchestration server-side onto your own GPUs.

Connects to: vLLM, OpenAI Responses API, Codex, MCP · · Apache-2.0 203★

Use it with an AI agent

Loadbay is an MCP server, so an agent can search the catalog and find this harness:

claude mcp add --transport http loadbay https://loadbay.xyz/api/mcp

• Source: https://github.com/vllm-project/agentic-api
• This harness as JSON: /api/harnesses/vllm-agentic-api
• Agent setup: /setup.md
• Browse all 700+ harnesses on Loadbay