Metasearch that queries your silos without moving a byte

What it does

Swirl is a Django-based metasearch engine that fans out one query to anything with a search API—SharePoint, Jira, BigQuery, GitHub, arXiv—and blends the results. It uses LLMs to re-rank the unified stream and can generate RAG answers with cited sources. A companion MCP server exposes the whole system to Claude, Cursor, and other agents, so they can query private sources without custom plugins.

The interesting bit

The system never indexes or extracts your data; it treats existing APIs and databases as the index. That eliminates ETL pipelines and stale copies, but it also means every query lives or dies by the latency and availability of the upstream source. The MCP adapter is the practical win: wire Swirl to your sources once, and every MCP-compatible agent inherits access.

Key highlights

• 40+ pre-configured connectors including M365, Snowflake, MongoDB, Elasticsearch, Pinecone, and ServiceNow
• LLM re-ranking of federated results and optional generative AI summaries with source citations
• Full REST API plus a Model Context Protocol adapter for agent integration
• Tag-prefix query routing (e.g., sharepoint:, github:) to target specific providers
• Docker-based evaluation environment; persistent deployments supported via Kubernetes

Caveats

• The Docker quickstart does not retain data or configuration between restarts, so it is strictly for evaluation
• RAG and AI summaries require an OpenAI or Azure OpenAI API key configured in Swirl’s environment
• Microsoft 365 connectors need OAuth2 app registration and administrator approval to function

Verdict

A solid fit if you need unified search across scattered enterprise sources but want to avoid another data migration. Less useful if you already have a centralized index and just need a new frontend.