MCPJungle Self-Hosted MCP Server Registry for Enterprise AI

Overview

MCPJungle is an open-source, self-hosted registry and gateway for Model Context Protocol (MCP) servers. It gives organizations a single, centrally managed MCP endpoint that aggregates and proxies multiple downstream MCP servers so AI agents and LLM clients can discover and call tools securely and consistently. Run inside your datacenter or cloud account to retain control over telemetry, access policies, and private integrations.

Repository: https://github.com/mcpjungle/MCPJungle

At runtime MCPJungle exposes a streamable HTTP MCP gateway (typically at /mcp) that MCP clients (Claude, Cursor, Copilot, etc.) can connect to. Developers use the CLI to register, group, enable/disable, and monitor downstream MCP servers and the individual tools they expose.

Features

• Central registry and gateway for multiple MCP servers
• Streamable HTTP and STDIO proxying of MCP tool calls
• Lightweight, single binary with an accompanying Docker image
• CLI to register/deregister and manage servers and tools
• Global enable/disable and tool grouping for access control
• Enterprise telemetry via OpenTelemetry and audit logs
• Self-hosted design — full control over privacy and networking
• Handles stateful/long-lived connections and cold-start coordination

Installation / Configuration

Install the MCPJungle CLI/binary via Homebrew or download releases:

# Homebrew (macOS)
brew install mcpjungle/mcpjungle/mcpjungle

# or pull the Docker image
docker pull ghcr.io/mcpjungle/mcpjungle

Run the registry server with Docker Compose (recommended for local or small deployments):

curl -O https://raw.githubusercontent.com/mcpjungle/MCPJungle/refs/heads/main/docker-compose.yaml
docker compose up -d

Run the server directly (example):

# start server binary on host
mcpjungle server --port 8080

The gateway endpoint is normally available at: http://:8080/mcp

Registering an MCP server

Use the CLI to add a remote MCP server:

mcpjungle register --name context7 --url https://mcp.context7.com/mcp

List registered servers:

mcpjungle list

Remove a server:

mcpjungle deregister --name context7

Example: Claude MCP client configuration

Use an MCP-remote shim to point the client to MCPJungle:

{
  "mcpServers": {
    "mcpjungle": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "http://localhost:8080/mcp",
        "--allow-http"
      ]
    }
  }
}

Available Resources

• CLI commands: register, deregister, list, enable/disable tools, export/import registry
• Registry web UI (when running Docker Compose) for visualization and management
• OpenTelemetry hooks for tracing and observability
• Audit logs for enterprise compliance

Use Cases

• Centralized tool discovery for enterprise AI agents: expose internal knowledge-base, code-search, and domain-specific tools via a single MCP endpoint.
• Multi-tenant gateway: provide different tool groups per team or project and selectively enable/disable tools.
• Secure external integrations: register third-party MCP providers behind corporate network controls while keeping control of access and telemetry.
• Cold-start optimization: maintain and coordinate stateful connections to downstream MCP servers for long-running streaming tools (e.g., tool chains that require session affinity).

Concrete example:

1. DevOps runs MCPJungle in a VPC and registers an internal “code-search” MCP server.
2. Product teams configure Claude instances to use the MCPJungle /mcp endpoint.
3. Agents discover the “code-search__get-function-docs” tool and call it without each agent needing direct access to the internal server.

Enterprise & Security

• Access control: enable/disable tools and create logical tool groups to limit exposure.
• Observability: integrate with OpenTelemetry to capture traces and metrics for tool-call latency and errors.
• Self-hosting: keep sensitive tools and logs inside your network; MCPJungle acts as the single controlled ingress for client tool calls.

Limitations & Notes

• Designed as a gateway/registry — it proxies and aggregates; downstream servers still host tool implementations.
• Cold-starts and long-lived streaming sessions require coordinating stateful connections; plan resource allocation accordingly.
• On macOS, prefer Homebrew for installation due to notarization constraints on distributed binaries.

Next Steps

• Clone the repo and try the quickstart Docker Compose setup.
• Register a public MCP (e.g., context7) via the CLI to experiment with tool calls from Claude or Cursor.
• Enable OpenTelemetry and review audit logs to integrate MCPJungle into your observability pipeline.

GitHub: https://github.com/mcpjungle/MCPJungle Tags: ai agents, ai-ml, infrastructure, llmops, mcp, mcp gateway, mcp registry, mcp server, model context protocol, self hosted