ragtime: Self-Host a RAG API and MCP Server

Overview

Ragtime is a self-hosted RAG service that brings local knowledge and configured tools into existing LLM clients. It combines an OpenAI-compatible chat API, an MCP server, and a built-in chat UI, with FAISS or PostgreSQL pgvector as vector-store options.

It is aimed at people who want to connect indexed files, repositories, databases, and other tools to coding assistants or OpenAI-compatible clients without relying on a separate hosted RAG service. The project supports hosted and local model providers, and also includes workspace previews for agent-driven development.

Key Features

• OpenAI-compatible /v1/chat/completions endpoint with streaming for compatible clients.
• MCP tools exposed over HTTP Streamable or stdio transports, with configurable authentication.
• Built-in chat UI with tool visualization, charts, tables, and sandboxed HTML components.
• Index uploads, Git repositories, and filesystems with FAISS or pgvector; schema and PDM indexes use pgvector.
• Configure hosted services and local runtimes including OpenAI, Anthropic, Ollama, llama.cpp, and LM Studio.
• Workspaces combine files, conversations, tools, and isolated runtime previews.
• Tool safeguards include SQL allowlist patterns, query LIMIT enforcement, and optional write-operation controls.

Use Cases

• A self-hosting team can index internal documentation or repositories and query them from an OpenAI-compatible client.
• Developers using Claude Desktop, VS Code Copilot, Cursor, or JetBrains IDEs can expose configured tools and knowledge through MCP.
• Teams with local model runtimes can connect indexed knowledge to those models while keeping the service and indexes on their own infrastructure.
• An organization working with databases or business systems can let assistants query configured tools, subject to its own permissions and security controls.

Getting Started

The README's Compose setup requires Docker and Docker Compose. Create a .env file from .env.example, set credentials, then start the stack:

docker compose up -d

See the README for the Compose configuration, provider setup, and security guidance.

Alternatives

• rag-web-ui: RAG Web UI focuses on document ingestion and web-based question answering, while ragtime also exposes an OpenAI-compatible RAG API and MCP server.
• rag-agent-platform: rag-agent-platform combines retrieval with configurable agents and MCP tools in a Java and Next.js stack, while ragtime centers on RAG access through its API, MCP server, and chat UI.
• MaxKB: MaxKB emphasizes building agents with workflows and external tools, while ragtime focuses on serving local knowledge through RAG APIs, MCP, and chat.

Considerations

• The project is active, with regular commits and releases, but the measured contributor base is small and highly concentrated. That is a maintenance and continuity consideration for production adopters.
• The recommended deployment is a multi-service Docker Compose stack, including PostgreSQL with pgvector, a runtime service, and SearXNG. This is more operational work than adopting a standalone library.
• Tool access can include database queries, SSH, shell execution, and workspace operations. The README warns that read-only controls are guardrails, not complete isolation; deployments need carefully scoped credentials and network boundaries.
• Some capabilities require elevated container or host permissions. Workspace preview isolation can fall back to chroot when mount authority is unavailable, and changing runtime isolation modes can risk snapshot history.
• The project is MIT-licensed. Review the license and any obligations of the external services or models you configure.