mcpjungle/MCPJungle

brew install mcpjungle/mcpjungle/mcpjungle
docker pull ghcr.io/mcpjungle/mcpjungle
curl -O https://raw.githubusercontent.com/mcpjungle/MCPJungle/refs/heads/main/docker-compose.yaml && docker compose up -d

Usage

• •Start the server (default gateway at http://localhost:8080/mcp): `mcpjungle start` or docker compose
• •Register a server: `mcpjungle register --name context7 --url https://mcp.context7.com/mcp` (also supports stdio servers via JSON config with `command`/`args`)
• •Connect Claude Desktop via npx mcp-remote `http://localhost:8080/mcp --allow-http`; Cursor/Copilot use the URL directly
• •Manage tools and groups: `mcpjungle list tools`, `mcpjungle disable tool context7__get-library-docs`, `mcpjungle create group -c ./group.json`

Key features

• ✓One MCP endpoint over streamable HTTP for Claude, Cursor, Copilot, Codex, and custom agents
• ✓Tool Groups: expose only a subset of tools to a client via a group-specific endpoint
• ✓Stateful connections (session_mode: stateful) to avoid cold-start latency on stdio servers
• ✓Enterprise mode: access control per MCP client, user accounts, and Prometheus-compatible OpenTelemetry metrics
• ✓Supports stdio and streamable-http upstream servers, plus enabling/disabling tools and prompts

Best for

Developers or teams who want to centralize and govern many MCP servers behind one endpoint without per-client config duplication.

Caveats

• ⚠macOS binary is not notarized yet — Homebrew install is required on macOS
• ⚠SSE (server-sent events) upstream support exists but is 'not currently mature'
• ⚠OAuth flow for authentication is not supported yet (static bearer tokens only)
• ⚠Tool Groups cannot be updated in place — you must delete and recreate; prompts are not supported in Tool Groups
• ⚠In enterprise mode, only an admin can create Tool Groups
• ⚠stdio servers need the larger `latest-stdio` Docker image (default docker-compose.yaml already uses it); other runtimes require a custom image