MCPJungle (MCP server) — Use Cases, Install & Live Demo

Name: MCPJungle MCP Server
Author: mcpjungle

Why use it

Key features

Single Go binary, no deps
Tool namespacing across upstream servers
stdio and Streamable HTTP transports
Hot-reload of registered upstreams
Tiny memory footprint (~20 MB)
MIT licensed

Live Demo

What it looks like in practice

mcpjungle-mcp.replay ▶ ready

0/0

Install

Pick your client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "mcpjungle-mcp": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "-p",
        "9090:9090",
        "ghcr.io/mcpjungle/mcpjungle:latest"
      ]
    }
  }
}

Open Claude Desktop → Settings → Developer → Edit Config. Restart after saving.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "mcpjungle-mcp": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "-p",
        "9090:9090",
        "ghcr.io/mcpjungle/mcpjungle:latest"
      ]
    }
  }
}

Cursor uses the same mcpServers schema as Claude Desktop. Project config wins over global.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "mcpjungle-mcp": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "-p",
        "9090:9090",
        "ghcr.io/mcpjungle/mcpjungle:latest"
      ]
    }
  }
}

Click the MCP Servers icon in the Cline sidebar, then "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "mcpjungle-mcp": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "-p",
        "9090:9090",
        "ghcr.io/mcpjungle/mcpjungle:latest"
      ]
    }
  }
}

Same shape as Claude Desktop. Restart Windsurf to pick up changes.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "mcpjungle-mcp",
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "-p",
        "9090:9090",
        "ghcr.io/mcpjungle/mcpjungle:latest"
      ]
    }
  ]
}

Continue uses an array of server objects rather than a map.

~/.config/zed/settings.json

{
  "context_servers": {
    "mcpjungle-mcp": {
      "command": {
        "path": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "-p",
          "9090:9090",
          "ghcr.io/mcpjungle/mcpjungle:latest"
        ]
      }
    }
  }
}

Add to context_servers. Zed hot-reloads on save.

claude mcp add mcpjungle-mcp -- docker run --rm -i -p 9090:9090 ghcr.io/mcpjungle/mcpjungle:latest

One-liner. Verify with claude mcp list. Remove with claude mcp remove.

Use Cases

Real-world ways to use MCPJungle

How to consolidate 8 MCP entries into one client config line

👤 Power users with many MCPs in Claude Desktop / Cursor ⏱ ~20 min intermediate

When to use: Your claude_desktop_config.json is 200 lines and editing it hurts.

Prerequisites

MCPJungle running locally — docker run -p 9090:9090 ghcr.io/mcpjungle/mcpjungle:latest or download a binary

Flow

Register upstreams

Register github, postgres, filesystem, fetch, and time as upstream MCPs in MCPJungle.✓ Copied

→ 5 upstreams listed; tool_namespaces show them
Switch client

Replace my 5 entries in claude_desktop_config.json with one entry pointing at MCPJungle on localhost:9090.✓ Copied

→ Single MCP entry; Claude lists all 5 servers' tools through it
Verify routing

Run a github tool and a postgres tool — confirm both reach their respective upstreams.✓ Copied

→ Both succeed; logs show correct routing

Outcome: Cleaner client config, identical capability, same latency.

Pitfalls

Two upstreams expose tools with the same name — MCPJungle prefixes by upstream name automatically — github.search_repositories vs gitlab.search_repositories

Run a shared team MCPJungle so devs share one curated set of MCPs

👤 Small teams (2–10 devs) ⏱ ~60 min advanced

When to use: Everyone keeps installing slightly different MCP versions; one team endpoint stops the drift.

Prerequisites

A shared host (small VM or shared dev box) — Run mcpjungle in Docker; expose port 9090 over Tailscale or VPN

Flow

Stand up the shared instance

Deploy MCPJungle on the team VM via docker compose. Bind to 100.x.x.x:9090 (Tailscale).✓ Copied

→ Service responds at the team URL
Pin upstream versions

Add github + filesystem + postgres MCPs at specific versions. Save the registration as a YAML file in the team repo.✓ Copied

→ registered.yaml committed; same on everyone's machine

Outcome: Whole team uses the same MCP versions and tools.

Pitfalls

Shared MCP shares creds — anyone using the endpoint can use them — Don't put write-capable MCPs in a shared instance unless you accept that

Combine with: github

Add and remove MCPs without restarting Claude

👤 Devs experimenting with new MCP servers ⏱ ~5 min beginner

When to use: You want to try a new MCP without exiting Claude and editing config.

Flow

Hot-add

Register a new MCP arxiv upstream into MCPJungle without restarting.✓ Copied

→ Tools appear in next Claude session refresh
Hot-remove

Unregister the slack MCP, since I'm done with that experiment.✓ Copied

→ slack tools no longer in tool list

Outcome: Faster MCP iteration loop.

Pitfalls

Some clients cache the tool list — Reload the conversation in Claude Desktop to refresh

Combinations

Pair with other MCPs for X10 leverage

mcpjungle-mcp + github

Manage the registered.yaml as code so the team config is version-controlled

Sync MCPJungle's registered upstreams from /infra/mcp/registered.yaml in the repo.✓ Copied

mcpjungle-mcp + filesystem

Persist boot-time registrations

Save the current upstream list to ./mcpjungle.json on shutdown.✓ Copied

Tools

What this MCP exposes

Tool	Inputs	When to call	Cost
register	name, transport: 'stdio'\|'http', command_or_url, env	Add an upstream MCP	free
unregister	name	Remove an upstream MCP	free
list_upstreams	(none)	See what's registered	free
<upstream>.<tool>	varies	Call any tool from any registered upstream	1 upstream call

Cost & Limits

What this costs to run

API quota: None (proxies upstream)
Tokens per call: Tiny overhead
Monetary: Free (MIT)
Tip: Run on the smallest box that fits your upstreams' memory needs

Security

Permissions, secrets, blast radius

Credential storage: Env vars passed to upstream subprocesses

Data egress: Same as upstream MCPs

Don't expose the gateway to the public internet without auth — wrap with mcpo or a reverse proxy if needed

Troubleshooting

Common errors and fixes

Upstream registered but no tools shown

MCPJungle proxies tool listing once on register — restart the upstream may be required if it dynamically declared tools after start

Verify: Run the upstream alone with `mcptools list` to confirm it lists tools

Stdio upstream zombie

MCPJungle reaps subprocesses on unregister — if a server is orphaned, restart MCPJungle

Verify: ps aux | grep <server>

Client says 'tool name conflict'

MCPJungle namespaces by upstream — call <upstream>.<tool>, not bare <tool>

Alternatives

MCPJungle vs others

Alternative	When to use it instead	Tradeoff
Archestra	You need SSO, audit, and policy on top of aggregation	Archestra is enterprise; MCPJungle is dev-weight
samanhappy/mcphub	You want a richer web UI	MCPHub has a UI; MCPJungle is purely binary + API

More

Resources

📖 Read the official README on GitHub

🐙 Browse open issues

🔍 Browse all 400+ MCP servers and Skills