Deployment¶

Deployment Options¶

salesforce-agent supports local stdio, a loopback-only development listener, a least-privilege stdio container, and a remote authenticated HTTPS boundary. Provider endpoint, credential, selector, identity, and trust material are supplied at runtime through AgentConfig; none is stored in this repository.

Installed stdio process¶

{
  "mcpServers": {
    "salesforce": {
      "command": "salesforce-mcp",
      "args": [],
      "env": {"MCP_TOOL_MODE": "intent"}
    }
  }
}

Loopback development listener¶

salesforce-mcp --transport streamable-http --host 127.0.0.1 --port 8000

Do not expose this listener beyond loopback. Network deployments require direct TLS or an explicitly trusted TLS-terminating ingress, configured authentication, exact MCP_ALLOWED_HOSTS, and an exact trusted-proxy CIDR policy.

Least-privilege local container¶

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 \
  registry.example.invalid/salesforce-agent@sha256:<digest> salesforce-mcp

The operator projects the selected AgentConfig profile into the process at runtime; the image remains immutable and contains no environment connection profile.

Remote authenticated HTTPS endpoint¶

{
  "mcpServers": {
    "salesforce": {"url": "https://service.example.invalid/mcp"}
  }
}

Store the real remote URL, outbound identity reference, and TLS-profile reference in AgentConfig, not in MCP client JSON or documentation.

MCP server transports¶

stdio (default)streamable-httpsse

salesforce-mcp

For local agent integration — the MCP client owns the process and speaks JSON-RPC over stdin/stdout.

salesforce-mcp --transport streamable-http --host 0.0.0.0 --port 8000

For networked deployments behind a reverse proxy. The server exposes /mcp for clients and /health for orchestrator checks.

salesforce-mcp --transport sse --host 0.0.0.0 --port 8000

Server-sent-events transport for clients that require it.

Health check¶

curl -fsS http://localhost:8000/health
# {"status": "OK"}

Docker Compose (MCP only)¶

cp .env.example .env   # fill in one auth flow
docker compose -f docker/mcp.compose.yml up -d

The MCP server listens on port 8000 (streamable-http) with a /health check.

Docker Compose (MCP + Agent)¶

docker compose -f docker/agent.compose.yml up -d

This brings up both the salesforce-agent-mcp service (port 8000) and the salesforce-agent-agent A2A service (port 9020, AG-UI web interface).

Building the image¶

docker build -f docker/Dockerfile -t example/salesforce-agent:agent-local .

A docker/debug.Dockerfile is provided for an in-place editable install with shell tooling and the Starship prompt.

A2A agent server¶

salesforce-agent                  # standalone A2A server

The agent connects to the MCP server via MCP_URL (http://salesforce-agent-mcp:8000/mcp in Compose) and exposes the A2A endpoint and AG-UI web interface on its port.

Environment¶

All configuration is via SALESFORCE_* environment variables — see .env.example. Mount secrets (client secret, JWT private key) from your secret store; never bake them into the image. Keep SALESFORCE_ALLOW_DESTRUCTIVE=False in shared deployments. Per-domain tool toggles (SOQLTOOL, RECORDSTOOL, DESCRIBETOOL, BULKTOOL, ADMINTOOL) control which tools are registered.

MCP client wiring¶

mcp_config.json at the repo root is the reference client entry (uv run salesforce-mcp plus the full environment block).

Reverse proxy + DNS (Caddy + Technitium)¶

For fleet deployments, publish the MCP server behind Caddy and register the hostname in Technitium DNS:

salesforce-mcp.example.invalid {
    reverse_proxy salesforce-agent-mcp:8000
}

Point an A record for salesforce-mcp.example.invalid at the ingress node in Technitium, then use https://salesforce-mcp.example.invalid/mcp as the client MCP_URL and https://salesforce-mcp.example.invalid/health as the health-check target.