Skip to main content
This page covers the day-to-day CLI commands used after the server is running: registering upstream MCP servers, inspecting what they expose, invoking tools, reading prompts and resources, and controlling visibility. Use the dedicated guides for deeper workflows:
Treat mcpjungle <command> --help as the authoritative source for the currently supported flags. This page keeps the higher-value workflows and examples.

register

Registers an upstream MCP server in mcpjungle. Once registration succeeds, the server’s tools, prompts, and resources become available through the gateway.
For streamable HTTP servers, you can use flags directly:
You can also pass a static bearer token for upstream authentication:
For stdio and sse servers, use a JSON config file:

Common flags

--name
string
Unique server name. Required when not using --conf.
--url
string
URL of a streamable HTTP MCP server. Required when registering an HTTP server without --conf.
--description
string
Optional human-readable description.
--bearer-token
string
Static bearer token forwarded to the upstream HTTP server.
--force
boolean
default:"false"
Deregister an existing server with the same name, then register the new one.
-c / --conf
string
Path to a JSON config file. Required for stdio and sse servers.

Minimal config examples

Streamable HTTP:
STDIO:
For the full config fields, session_mode, environment-variable placeholders, and transport-specific notes, use:
Server names must be unique across mcpjungle and must not contain whitespace, special characters, or consecutive underscores (__).

deregister

Removes a registered MCP server and all of its tools, prompts, and resources from the gateway.
Example:
Deregistering a server removes its tools, prompts, and resources immediately. Any connected MCP client relying on them will start failing.

list

Lists the entities currently registered in mcpjungle.

list servers

Lists registered MCP servers and their transport details.

list tools

Lists tools across the gateway, or filters them by server or tool group.
--server
string
Filter tools to a specific registered server.
--group
string
Filter tools to those currently active in a tool group.
--server and --group cannot be combined. list tools --group only shows tools that are currently active. Use get group to inspect the full configured group definition.

list prompts

Lists prompts across all servers or from one server.

list resources

Lists resources across all servers or from one server.

list groups

Lists tool groups.

Common examples

usage

Shows the input schema for a tool before you invoke it.
Example:
The output includes the tool description, argument types, required fields, and any JSON-schema constraints exposed by the upstream MCP server.

invoke

Calls a tool directly through mcpjungle.

Flags

--input
string
default:"{}"
JSON object of input arguments passed to the tool.
--group
string
Invoke the tool within a tool group’s context. The tool must be available in that group.

Examples

Text responses are printed to stdout. If the response contains images, audio, or binary resources, the CLI may save files in the current working directory.

enable

Enables tools, prompts, or an entire server globally.
Supported forms:
  • mcpjungle enable tool <name>
  • mcpjungle enable prompt <name>
  • mcpjungle enable server <name>
Examples:
The legacy form mcpjungle enable <name> still works for backward compatibility, but enable tool <name> or enable server <name> is preferred.

disable

Disables tools, prompts, or an entire server globally.
Supported forms:
  • mcpjungle disable tool <name>
  • mcpjungle disable prompt <name>
  • mcpjungle disable server <name>
Examples:
The legacy form mcpjungle disable <name> still works for backward compatibility, but only maps to tool disabling behavior.

get prompt

Retrieves a prompt template and optionally renders it with arguments.
Example:
The output shows the generated structured messages returned by the upstream MCP server.

get resource

Retrieves resource metadata, or reads the resource contents.
Examples:
With --read, text content is printed inline. Blob content may be decoded and written to a file in the current working directory.