Swirl Search

At a Glance:

Swirl is an open-source federated search and RAG platform that distributes queries to search APIs, databases, and enterprise sources, re-ranks results with LLMs without moving data, and exposes the unified search via REST API and Model Context Protocol.

Overview:

Swirl is a federated metasearch and retrieval-augmented generation platform that adapts user queries and distributes them to anything with a search API — search engines, databases, NoSQL engines, cloud services, and data silos. It uses Large Language Models to re-rank the unified results without extracting or indexing any data from the connected sources. Swirl exposes its search and RAG capabilities through a REST API and via the Model Context Protocol, allowing MCP clients like Claude Desktop and Cursor to answer questions against private sources. The platform ships with over 40 pre-configured connectors covering public web, scientific literature, Microsoft 365, developer tools, CRMs, databases, and vector stores.

Key Decision Points:

• Federated architecture with no data movement: Swirl queries sources in place and re-ranks results with LLMs without extracting or indexing data, which matters for environments where data locality and access controls must be preserved.

• MCP server integration available: A separate swirl-mcp-server exposes Swirl's search and RAG as MCP tools, making private enterprise sources queryable from Claude Desktop, Claude Code, Cursor, and other MCP clients.

• Docker quickstart intended for evaluation only: The one-command Docker path does not persist data or configuration when the container shuts down. Persistent deployment requires following the installation guide or Kubernetes guide.

• Requires credentials for many connectors: SearchProvider configurations, Microsoft 365 OAuth2, and AI/RAG providers all require credential setup. M365 setup requires Azure App Registration and M365 administrator approval.

• Query routing via tag-prefix syntax: Users can scope searches to specific sources using prefix shortcuts like sharepoint:, onedrive:, arxiv:, and github:, which route the query to only matching providers.

Core Features:

• Federated query distribution: Adapts and distributes user queries to multiple search providers simultaneously, handling per-provider query adaptation including NOT rewriting and AND/OR pass-through.

• Microsoft 365 integration with OAuth2: Searches across Outlook, OneDrive, SharePoint, and Teams using OAuth2 authentication, with tag-prefix shortcuts to scope queries to individual M365 sources.

• LLM-based re-ranking and RAG: Uses AI models like ChatGPT to re-rank unified results and generate RAG answers without extracting or indexing source data.

• Model Context Protocol server: The swirl-mcp-server exposes six MCP tools — search, rag_answer, list_providers, create_search, get_results, and list_searches — for integration with MCP-compatible AI tools.

• 40+ pre-configured connectors: Ships with SearchProviders for Google PSE, arXiv, GitHub, Jira, Confluence, HubSpot, BigQuery, Snowflake, MongoDB, Elasticsearch, Pinecone, Qdrant, and more.

• Configurable result processing pipeline: Supports pipelining of Processor stages for real-time query, response, and result transformation, plus multiple result mixers including relevancy, date, and round-robin.

Use Cases:

• Developers and teams that need to search across many internal sources without moving data: Swirl queries databases, document stores, and SaaS tools in place, returning unified and re-ranked results through a single API.

• AI tool users who want private data available in their MCP-compatible clients: Through swirl-mcp-server, users can make their connected enterprise sources queryable from Claude Desktop, Claude Code, or Cursor without per-source agent plumbing.

• Organizations with Microsoft 365 that want federated search across SharePoint, OneDrive, Outlook, and Teams: Swirl supports OAuth2-based M365 search with tag prefixes for scoping queries to specific M365 sources.

Open-Source Alternative Value:

Swirl provides a self-contained federated search and RAG layer that connects to existing data sources without requiring data extraction or indexing — an architectural approach that keeps data in place while making it searchable through a unified API. The integration with the Model Context Protocol through swirl-mcp-server means AI tools can query private enterprise sources without individual connector setup in each agent. With 40+ pre-configured SearchProviders and an extensible object model for connectors and mixers, the platform offers a starting point for organizations that need federated search across heterogeneous internal and external sources, backed by LLM re-ranking and RAG.