Deployment¶

Deployment Options¶

emerald-exchange exposes its MCP server (console script emerald-exchange-mcp) four ways. Pick the row that matches where the server runs relative to your MCP client, then copy the matching mcp_config.json below. Add the service-connection environment variables documented in the Configuration section. Provider endpoint, credential, selector, identity, and trust material are supplied at runtime through AgentConfig (~/.config/agent-utilities/config.json) or environment — none of it is stored in this repository or belongs hardcoded in mcp_config.json.

#	Option	Transport	Where it runs	`mcp_config.json` key
1	stdio	`stdio`	client launches a subprocess	`command`
2	Streamable-HTTP (local)	`streamable-http`	a local network port	`command` or `url`
3	Local container / uv	`stdio` or `streamable-http`	Docker / Podman / uv on this host	`command` or `url`
4	Remote URL	`streamable-http`	a remote host behind Caddy	`url`

1. stdio (local subprocess)¶

The client launches the server over stdio via uvx — best for local IDEs (Cursor, Claude Desktop, VS Code):

{
  "mcpServers": {
    "emerald-exchange-mcp": {
      "command": "uvx",
      "args": ["--from", "emerald-exchange", "emerald-exchange-mcp"],
      "env": {"MCP_TOOL_MODE": "condensed"}
    }
  }
}

2. Streamable-HTTP (local process)¶

Run the server as a long-lived HTTP process:

uvx --from emerald-exchange emerald-exchange-mcp --transport streamable-http --host 0.0.0.0 --port 8000
curl -s http://localhost:8000/health        # {"status":"OK"}

For a loopback-only development listener (no published port, nothing else on the host or network can reach it), bind --host 127.0.0.1 instead of 0.0.0.0. Do not expose a 0.0.0.0 listener beyond a trusted network boundary: a real network deployment needs direct TLS or an explicitly trusted TLS-terminating ingress, configured authentication, an exact MCP_ALLOWED_HOSTS, and an exact trusted-proxy CIDR policy (see Behind a Caddy reverse proxy below).

Then either let the client launch it:

{
  "mcpServers": {
    "emerald-exchange-mcp": {
      "command": "uvx",
      "args": ["--from", "emerald-exchange", "emerald-exchange-mcp", "--transport", "streamable-http", "--port", "8000"],
      "env": {
        "TRANSPORT": "streamable-http",
        "HOST": "0.0.0.0",
        "PORT": "8000"
      }
    }
  }
}

…or connect to the already-running process by URL:

{
  "mcpServers": {
    "emerald-exchange-mcp": { "url": "http://localhost:8000/mcp" }
  }
}

3. Local container / uv¶

(a) Launch a container directly from mcp_config.json (stdio over the container — no ports to manage). Swap docker for podman for a daemonless runtime:

{
  "mcpServers": {
    "emerald-exchange-mcp": {
      "command": "docker",
      "args": [
        "run", "-i", "--rm",
        "-e", "TRANSPORT=stdio",
        "knucklessg1/emerald-exchange:2.1.0"
      ]
    }
  }
}

(a2) Least-privilege container run — the same launch, hardened for an untrusted or shared host (read-only root filesystem, all capabilities dropped, no privilege escalation, a bounded process count, and a noexec tmpfs for the one writable path the process needs):

docker run -i --rm \
  --read-only \
  --cap-drop=ALL \
  --security-opt=no-new-privileges \
  --pids-limit=256 \
  --tmpfs /tmp:rw,noexec,nosuid,nodev,size=64m \
  -e TRANSPORT=stdio \
  knucklessg1/emerald-exchange@sha256:<digest> emerald-exchange-mcp

Pin the immutable release digest (rather than the mutable floating tag) for this form, and project the selected AgentConfig profile into the process at runtime — the image itself carries no environment-specific connection profile.

(b) Run a local streamable-http container, then connect by URL:

docker run -d --name emerald-exchange-mcp -p 8000:8000 \
  -e TRANSPORT=streamable-http \
  -e PORT=8000 \
  knucklessg1/emerald-exchange:2.1.0
# or, from a clone of this repo:
docker compose -f docker/mcp.compose.yml up -d

{
  "mcpServers": {
    "emerald-exchange-mcp": { "url": "http://localhost:8000/mcp" }
  }
}

(c) From a local checkout with uv:

uv run emerald-exchange-mcp --transport streamable-http --port 8000

4. Remote URL (deployed behind Caddy)¶

When the server is deployed remotely (e.g. as a Docker service) and published through Caddy at a deployment-selected HTTPS hostname, connect with the "url" key — no local process or image required:

{
  "mcpServers": {
    "emerald-exchange-mcp": { "url": "https://emerald-exchange-mcp.example.invalid/mcp" }
  }
}

Caddy reverse-proxies https://emerald-exchange-mcp.example.invalid to the container's :8000 streamable-http listener; https://emerald-exchange-mcp.example.invalid/health returns {"status":"OK"} when the service is live. Keep the real remote URL, outbound identity references, and TLS trust profile in AgentConfig (~/.config/agent-utilities/config.json) rather than duplicating them across mcp_config.json files or documentation.

This page covers running emerald-exchange as a long-lived service: the transports, the companion A2A agent server, a Docker Compose stack, putting it behind a Caddy reverse proxy, and giving it a DNS name with Technitium.

emerald-exchange ships two console scripts: the MCP server (emerald-exchange-mcp) and a full A2A agent server (emerald-exchange-agent) that drives the MCP toolset through a Pydantic-AI agent. A text-mode cockpit (emerald-cockpit) is documented in Usage.

Run the MCP server¶

The transport is selected with --transport (or the TRANSPORT env var):

stdio (default)streamable-httpsse

emerald-exchange-mcp

For IDE / desktop MCP clients that launch the server as a subprocess.

emerald-exchange-mcp --transport streamable-http --host 0.0.0.0 --port 8100

A network server with a /health endpoint and /mcp route.

emerald-exchange-mcp --transport sse --host 0.0.0.0 --port 8100

Health check (HTTP transports):

curl -s http://localhost:8100/health        # {"status":"OK"}

Configuration (environment)¶

The MCP server starts with no required configuration — the default Paper backend is fully self-contained. Live trading and data connectors read their own credentials and remain inactive when those credentials are absent. The variables most commonly set:

Var	Default	Meaning
`TRANSPORT`	`stdio`	Transport: `stdio`, `streamable-http`, or `sse`
`HOST`	`127.0.0.1`	Bind address for HTTP transports
`PORT`	`8100`	Listen port for HTTP transports
`FUNDAMENTALSTOOL`	`True`	Register the SEC EDGAR fundamentals tool group
`EDGAR_IDENTITY`	—	SEC identity (`"Name email@example.com"`) for fundamentals
`WALLETINTELTOOL`	`True`	Register the Polymarket wallet-intelligence tool group
`POLY_TRADES_PATH`	—	Path to a processed Polymarket trades CSV/Parquet
`ALPACA_API_KEY` / `ALPACA_SECRET_KEY`	—	Alpaca credentials (only when using Alpaca)
`BINANCE_API_KEY` / `BINANCE_SECRET`	—	Binance credentials (only when using Binance)
`COINBASE_API_KEY` / `COINBASE_SECRET`	—	Coinbase credentials (only when using Coinbase)

All trading behavior — the default backend, default mode, risk limits, and per-exchange settings — is configured in the trading block of ~/.config/agent-utilities/config.json. The full schema, the backend matrix, and the complete environment-variable list are documented in the Configuration schema. See also Configuration, trust, and privacy for the runtime-only-secrets policy this package follows.

Backing service¶

emerald-exchange connects to managed, SaaS exchange and data APIs — Alpaca, Binance, Coinbase, Kraken (via CCXT), Kalshi, Polymarket, and SEC EDGAR. These are provider-hosted services with no local deployment, so there is no backing-system recipe to provision: only connection configuration (API keys and identities) is required, and the package operates entirely against the in-process Paper backend until those credentials are supplied.

Docker Compose¶

The repository ships docker/mcp.compose.yml. It reads a sibling .env, publishes the HTTP MCP server on :8100, and starts the companion agent server on :9100:

services:
  emerald-exchange-mcp:
    image: knucklessg1/emerald-exchange:2.1.0
    container_name: emerald-exchange-mcp
    hostname: emerald-exchange-mcp
    command: ["emerald-exchange-mcp"]
    restart: always
    env_file:
      - ../.env
    environment:
      - PYTHONUNBUFFERED=1
      - HOST=0.0.0.0
      - PORT=8100
      - TRANSPORT=streamable-http
    ports:
      - "8100:8100"
    healthcheck:
      test: ["CMD", "python3", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:8100/health')"]
      interval: 30s
      timeout: 10s
      retries: 3

cp .env .env.local            # then edit the connector credentials you use
docker compose -f docker/mcp.compose.yml up -d
docker compose -f docker/mcp.compose.yml logs -f

A2A agent server¶

The agent server (emerald-exchange-agent) launches emerald-exchange as a full Pydantic-AI agent with Agent-to-Agent support, wired to the MCP toolset via MCP_URL. It is defined alongside the MCP server in docker/mcp.compose.yml and listens on :9100:

  emerald-exchange-agent:
    image: knucklessg1/emerald-exchange:2.1.0
    container_name: emerald-exchange-agent
    hostname: emerald-exchange-agent
    command: ["emerald-exchange-agent"]
    depends_on:
      - emerald-exchange-mcp
    restart: always
    env_file:
      - ../.env
    environment:
      - PYTHONUNBUFFERED=1
      - HOST=0.0.0.0
      - PORT=9100
      - MCP_URL=http://emerald-exchange-mcp:8100/mcp
      - PROVIDER=openai
      - LLM_BASE_URL=${LLM_BASE_URL:-http://host.docker.internal:1234/v1}
      - LLM_API_KEY=${LLM_API_KEY:-llama}
      - MODEL_ID=${MODEL_ID:-qwen/qwen3.5-9b}
      - ENABLE_WEB_UI=True
    ports:
      - "9100:9100"

Run it directly:

emerald-exchange-agent --host 0.0.0.0 --port 9100 \
  --mcp-url http://localhost:8100/mcp

Behind a Caddy reverse proxy¶

Expose the HTTP server on a hostname with automatic TLS. Add to your Caddyfile:

# Deployment-selected HTTPS hostname
emerald-exchange.example.invalid {
    tls internal
    reverse_proxy emerald-exchange-mcp:8100
}

# Public — automatic Let's Encrypt
emerald-exchange.example.com {
    reverse_proxy emerald-exchange-mcp:8100
}

Reload Caddy:

docker compose -f services/caddy/compose.yml exec caddy caddy reload --config /etc/caddy/Caddyfile

DNS with Technitium¶

Point the hostname at the host running Caddy. Via the Technitium API:

curl -s "https://dns-admin.example.invalid/api/zones/records/add" \
  --data-urlencode "token=$TECHNITIUM_DNS_TOKEN" \
  --data-urlencode "domain=emerald-exchange.example.invalid" \
  --data-urlencode "zone=arpa" \
  --data-urlencode "type=A" \
  --data-urlencode "ipAddress=10.0.0.10" \
  --data-urlencode "ttl=3600"

…or add an A record emerald-exchange.example.invalid → <caddy-host-ip> in the Technitium web console (https://dns-admin.example.invalid). The ecosystem technitium-dns-mcp automates this as a tool.

Register with an MCP client¶

Add to your client's mcp_config.json (multiplexer nickname ee):

{
  "mcpServers": {
    "emerald-exchange": {
      "command": "uv",
      "args": ["run", "emerald-exchange-mcp"],
      "env": {}
    }
  }
}

For a remote HTTP server, point the client at https://emerald-exchange.example.invalid/mcp instead.