Airweave
by airweave-ai
Airweave syncs 40+ apps and databases into one searchable index, then exposes that index to agents through an MCP server with instant, classic, and agentic search modes.
Run the MCP search server
npx -y airweave-mcp-search
About
Airweave sits between your data sources and your agents. You connect sources — Slack, GitHub, Notion, Google Drive, Confluence, and dozens more — into a collection; Airweave handles the OAuth, ingestion, chunking, and indexing so the agent side never has to touch a scraper or a vector database directly. Once a collection is synced, the MCP server exposes it as a single `search-{collection}` tool that any MCP-compatible client can call.
That search tool actually runs in one of three modes: instant vector search for sub-second lookups, classic mode where an LLM plans the retrieval strategy (the default, taking a few seconds), or agentic mode where the search itself is a multi-step exploration for harder questions. The server ships as a local stdio process for desktop clients or as a hosted Streamable HTTP endpoint, authenticated by API key or OAuth 2.0, so the same collection can back a laptop-based agent and a cloud deployment without re-plumbing.
Key features
- 40+ prebuilt connectors (Slack, GitHub, Notion, Google Drive, Confluence, and more)
- Three search modes — instant, classic (LLM-planned), and agentic — tradeable for speed vs. depth
- Single MCP tool per collection (`search-{collection}`) plus a `get-config` inspection tool
- Local stdio mode for desktop clients or hosted Streamable HTTP for cloud agents
- API-key or OAuth 2.0 authentication
- Continuous background sync so indexed data stays current without manual re-ingestion
Use cases
- Giving a coding agent unified search across a team's Slack, Jira, and Confluence without building a custom RAG pipeline
- Backing a support or research assistant with live, synced knowledge from multiple SaaS tools
- Standing up a self-hosted retrieval layer for internal documents that any MCP client can query
- Swapping data sources in and out of an agent's context without rewriting client-side integration code
Available tools
search-{collection}
Searches a specific synced collection; supports instant, classic, or agentic retrieval depth.
get-config
Returns the MCP server's current configuration and collection metadata.
Frequently asked questions
Do I need to self-host Airweave to use the MCP server?
No — you can point the MCP server at Airweave's hosted platform with an API key and collection ID, or self-host the full stack via Docker if you want to run it yourself.
What's the difference between the three search modes?
Instant is a direct vector lookup, classic lets an LLM plan the query strategy before searching (the default), and agentic runs a multi-step search for questions that need more exploration.