ratel

Context engineering layer for AI agents that cuts token usage by ~80% through in-process BM25 and semantic skill retrieval, progressive tool disclosure, and agent memory. Exposes an MCP server so any coding agent or LLM host can connect to it.

Connects to: Claude Code, MCP-compatible hosts, BM25 retrieval, semantic search, agent memory · · MIT 425★

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/ratel-ai/ratel
• This harness as JSON: /api/harnesses/ratel
• Agent setup: /setup.md
• Browse all 700+ harnesses on Loadbay