Each tenant gets its own MCP endpoint (`/api/v1/{tenant}/mcp`) with path-based isolation. Full MCP protocol support — Streamable HTTP, WebSocket, and SSE — with protocol version negotiation, resumable SSE streams, and JSON-RPC batching.

340 tools across 23 native connectors:

- Code & VCS: GitHub (24), GitLab (22), Bitbucket (19) - Project management: Jira (20), Linear (18), Confluence (16) - Communication: Slack (11), Discord (15), Microsoft Teams (13) - Email: Gmail (14), Outlook (14) - Documents: Google Docs (10), Sheets (14), Slides (11), Notion (10), Excel (14), PowerPoint (10), Zoom (12) - AI coding tool intelligence: GitHub Copilot (19), Claude Code (18), OpenAI Codex (16), Cursor (18), Windsurf (11) — standardized metrics schema for cross-tool comparison of adoption, cost, and productivity

Plus support for hosting any external MCP server (Python/Node.js/Go) as a stdio subprocess with health checks and auto-restart.

Under the hood: LRU server pooling (5K max instances, 30-min TTL), session management via `Mcp-Session-Id`, per-tenant token-bucket rate limiting, Prometheus metrics, and feature flags for progressive rollout.

Stack: FastAPI, React, SQLAlchemy, PostgreSQL/Supabase, Docker & Kubernetes with Helm charts.

GitHub: https://github.com/sagemcp/sagemcp License: Apache 2.0

Happy to answer questions about the architecture or multi-tenant MCP patterns.