What is mcprack?

The problem

Giving an AI client access to an MCP server normally means installing the server on every workstation, putting its credentials into a JSON file and keeping both up to date. That scales badly and leaks secrets into files that get copied, committed and backed up.

How mcprack solves it

AI client (Claude, Cursor, VS Code ...)
        │   HTTPS, one URL per server
        ▼
mcprack  /proxy/mcp/...        ← identifies the user, checks permissions
        │
        ├─ server has a command → per-user fastmcp process (loopback)
        └─ server has a URL     → relay to that backend
                                  (credentials injected on the server)

  1. An administrator registers MCP servers (a local command, or a network URL) and stores their credentials once.

  2. A user selects the servers they are allowed to use and downloads or syncs the generated configuration.

  3. The client sends every request to mcprack, which authenticates the caller, starts or reaches the backend and forwards the request.

Who uses what

Role

Typical tasks

Administrator

Register servers, store credentials, grant access, apply templates, read the audit trail, run mcprack CLI commands

User

Choose servers, set personal credential overrides, create API tokens, download or sync the client configuration

Operator

Install the package, configure /etc/mcprack/env, put a reverse proxy in front, monitor the service

What mcprack is not

  • It does not implement MCP servers; it hosts and relays them. Servers that ship a mcprack companion package register themselves — see MCP servers with a mcprack companion.

  • It is not an identity provider. It authenticates against local accounts or LDAP / Active Directory, and issues OAuth tokens only for MCP clients.