Argus MCP - a single gateway guarding your MCP servers

Connect any MCP client to all your MCP servers through a single server.

Argus MCP is a central gateway and management platform for the Model Context Protocol. Rather than configure each AI client to talk to each MCP server individually, point your clients at Argus and let it handle the rest - aggregating capabilities, routing tool calls, managing connections, and securing everything behind one endpoint.

Why use Argus MCP? One endpoint, many backends: MCP clients connect to one Argus address - Argus aggregates tools, resources, and prompts from all your backend servers. Stdio isolation: Run stdio MCP servers in isolated Docker containers, or deploy them on Kubernetes. Any transport: Backends can be local stdio processes, remote SSE endpoints, or Streamable HTTP servers. Argus works with all three. Secure by default: JWT/OIDC authentication, RBAC authorization, encrypted secret storage, and automatic log redaction. Live management: Hot-reload config, reconnect backends, and monitor everything through the REST API, TUI, or CLI/REPL. Runs anywhere: Install from PyPI, run in Docker, or build from source.

Components

Argus MCP has a server/client architecture:

Short alias: amcp is a convenience alias for argus-mcp (e.g. amcp server).

Companion packages:

Links

• Getting started
• Configuration reference
• Management API
• Docker guide
• Kubernetes guide
• Security guide
• CLI reference
• TUI guide
• Changelog
• Report an issue

How it works

1. Configure backends in a YAML file — stdio processes, SSE servers, or HTTP endpoints.
2. Start with argus-mcp server — Argus connects to all backends and aggregates their capabilities.
3. Point any MCP client (Claude Desktop, Cursor, Copilot, Cline, etc.) at the Argus endpoint.
4. Monitor via the TUI, REST API, or hot-reload config without restarting.

Argus exposes two transports simultaneously:

Installation

Docker (recommended — no Python required):

docker run --rm -p 9000:9000 diaz3618/argus-mcp:latest server --host 0.0.0.0

Published to Docker Hub and GHCR with multi-arch support (amd64 + arm64). See the Docker guide for Compose files and volume mounts.

PyPI (requires Python 3.10+):

uv tool install argus-mcp   # recommended
pipx install argus-mcp      # alternative

Try without installing: uv run argus-mcp --help

See getting started for source installation and Makefile shortcuts.

Quick start

# Start the server (auto-detects config.yaml)
argus-mcp server

# Detached mode
argus-mcp server --detach --name my-server
argus-mcp status
argus-mcp stop my-server

Config file resolution: --config flag → ARGUS_CONFIG env → auto-detect (config.yaml → config.yml).

See the getting started guide for a full walkthrough.

Configuration

A minimal config to get started:

version: "1"

backends:
  filesystem:
    type: stdio
    command: npx
    args: ["-y", "@modelcontextprotocol/server-filesystem", "/tmp"]

  remote_tools:
    type: sse
    url: "https://mcp.example.com/sse"

Backends can be stdio (local process), sse, or streamable-http. All string values support ${VAR_NAME} env expansion and secret:name references to the encrypted secret store.

See the configuration reference for all fields, backend types, auth options, RBAC policies, and feature flags.

Management API

A REST API at /manage/v1/ provides full runtime control: health/readiness probes, backend reconnection, hot-reload, live event streaming, capability proxying, and more.

Set ARGUS_MGMT_TOKEN to require Authorization: Bearer <token> on all management requests.

See the API reference for the full endpoint list.

TUI

argus-mcp tui --server http://127.0.0.1:9000

The TUI supports multi-server mode via a servers.json config — switch between servers from a single dashboard. See the TUI guide for details.

Contributing

• Report issues
• Pull requests

License

This project is licensed under the GNU General Public License v3.0.