Status: coming soon. MCP connector support is on the roadmap and not yet available in the product. The design below describes the planned integration.
Overview
feder8d is designed to speak the Model Context Protocol. Any MCP server you can run (file system, GitHub, Slack search, your own) becomes a retrieval source or a tool.
Setup
- In the Tenant Console, Connectors → MCP → Add server.
- Provide the stdio command or HTTP base URL.
- Configure auth (if HTTP). Stdio servers run in the connector worker pod scoped to your tenant.
- Mark which tools should be exposed to chat (default: none — explicit allowlist).
Tenant isolation
Each MCP server runs in a worker pod whose namespace is the tenant’s namespace. The connector process cannot see another tenant’s data even if the underlying MCP server is shared OSS code.
When to use MCP vs a native connector
- Native (Notion / Drive / Confluence): faster sync, indexed in Qdrant for retrieval at query time.
- MCP: live tool use at inference time, no pre-indexing required, but slower per query.
Use both: native connectors for the long tail of knowledge, MCP for live actions (create a Linear issue, run a script, post a Slack message).