🐰 Ragbits Claude Code Plugin

A Claude Code plugin that scaffolds ready-to-run ragbits applications from templates.

Homepage | Ragbits Documentation | Contact

---

Available Skills

/chat

Scaffold a ragbits chat application with a conversational web UI.

/chat my-chatbot --model gpt-4o --features auth,feedback,theme

Options:

• --model MODEL — LLM model via LiteLLM (default: gpt-4o-mini)
• --features — comma-separated: auth, agents, upload, feedback, theme, document-search

/agent

Scaffold a ragbits agent application — a tool-using assistant with the canonical Agent + Prompt + LiteLLM pattern. Also handles A2A servers, MCP tool integration, multi-agent orchestrators, and agent hooks (validation, logging, guardrails, confirmation).

/agent customer-support A bot that handles refund questions --model gpt-4.1

Options:

• --model MODEL — any LiteLLM-compatible model (default: gpt-4.1)

The skill infers features from the description — mention MCP, A2A, hooks, or orchestration and the relevant components are added automatically.

/rag

Scaffold a ragbits RAG application with document ingestion, vector search, and LLM-powered Q&A.

/rag my-kb --model gpt-4o --vector-store chroma --features chat-ui,reranker

Options:

• --model MODEL — LLM model for answer generation (default: gpt-4o-mini)
• --embedding-model MODEL — embedding model (default: text-embedding-3-small)
• --vector-store STORE — backend: in-memory, chroma, qdrant (default: in-memory)
• --features — comma-separated: chat-ui, auth, feedback, theme, reranker, rephraser

/evaluate

Scaffold a ragbits evaluation project that measures RAG / retrieval / answer quality using the ragbits-evaluate framework. Supports document-search precision/recall/F1, LLM-as-judge metrics for question-answering, synthetic dataset generation, and Optuna hyperparameter sweeps.

/evaluate retrieval-bench --target document-search --features optimize

Options:

• --target TARGET — document-search, question-answer, or custom (default: document-search)
• --judge-model MODEL — LLM for judge-style metrics (default: gpt-4o-mini)
• --embedding-model MODEL — embedding model (default: text-embedding-3-small)
• --features — comma-separated: optimize, dataset-generator, custom-metric, custom-pipeline

/mcp

Scaffold an MCP (Model Context Protocol) server that exposes a ragbits agent, RAG pipeline, or prompt to any MCP client (Claude Desktop, Claude Code, Cursor). Built as a bridge: the official mcp SDK (FastMCP) hosts the server, ragbits primitives power the tool bodies.

/mcp docs-mcp Expose our product documentation RAG to Claude Desktop --backend rag

Options:

• --backend BACKEND — agent, rag, prompt, or plain (inferred from the description by default)
• --transport TRANSPORT — stdio (default), sse, or streamable-http
• --model MODEL — LiteLLM-compatible model for LLM-backed tools (default: gpt-4.1)

Note: /agent covers the client direction (a ragbits agent calling an external MCP server). /mcp is the server direction — exposing ragbits capabilities to external MCP clients.

Installation

Add the plugin to your Claude Code project:

claude plugin marketplace add https://github.com/deepsense-ai/ragbits-cc-plugin.git
claude plugin install ragbits@ragbits-cc-plugin

License

Ragbits is licensed under the MIT License.