- Authenticated access to MCPs
- MCP clients must be created explicitly
- Each client’s access to MCPs is controlled through allow-lists
- Human users are separated from machine identities
Development mode versus enterprise mode
| Capability | Development mode | Enterprise mode |
|---|---|---|
| Gateway authentication | Disabled | Required |
| MCP client access control | Open access to registered servers | Explicit allow-list per client |
| User accounts | Not needed | Supported |
| OpenTelemetry metrics | Disabled by default | Enabled by default |
| Server initialization | Automatic | Manual initialization by an Administrator |
Governance model
Mcpjungle uses three main identity and access concepts in enterprise mode.Administrator
Admin initializes the server, register MCP servers, creates mcp clients, creates users, and manages platform configuration.MCP clients
MCP clients are machine identities for your AI clients (agents, CLIs like claude code, IDEs like Cursor, or custom mcp clients). Each client gets:- an access token
- an allow-list of upstream MCP servers it may access
Users
Users are human identities for people who need authenticated access to Mcpjungle. They have a narrower permission set than admins.Typical enterprise flow
Start Mcpjungle in enterprise mode
Deploy the gateway in enterprise mode and connect it to PostgreSQL.
Create MCP clients
Create one machine identity per client or workflow and grant access only to the servers that client should reach.
Why this matters
For a shared deployment, a serious product needs more than just a gateway endpoint. It needs a control boundary:- which client can reach which server
- which humans can operate the system
- how tokens are issued and rotated
- how usage can be observed centrally
Next steps
Upstream authentication
Configure bearer tokens and custom headers for Mcpjungle’s upstream server connections.
Access control
Create MCP clients and define exactly which servers they can access.
Clients and users
Manage machine identities and human accounts cleanly.
Production deployment
Run Mcpjungle with the expected enterprise operating model.
