Quickstart - Dome

Call MCP tools and LLM models through the Dome gateway from a Python application. The SDK handles authentication, act-as encoding, audit correlation, and typed error decoding — your code talks to one client.

Prerequisites:

Python 3.12+
A registered agent and an agent API key (dome_...) — refer to Developer Quickstart
The gateway URL for the Gateway the agent is granted (https://gateway.../gateways/{id})

Install

pip install dome-sdk

Provider SDKs are optional. client.gateway.openai_client() lazy-imports openai; client.gateway.anthropic_client() lazy-imports anthropic. Install those packages only in agents that use the provider-native factories.

Initialize

Construct a Client with the agent token and gateway URL. The SDK does not read environment variables — pass values directly.

import dome

client = dome.Client(
    token="{{AGENT_TOKEN}}",
    gateway_url="https://gateway.domesystems.ai/gateways/{{GATEWAY_ID}}",
    act_as_method="none",
)
client.connect()

The /gateways/{id} segment names the Gateway the agent is granted. Pass the complete URL returned by agent-key creation or token exchange. Bare roots and URLs ending in /mcp or /v1 raise DomeGatewayConfigurationError at connect() — the SDK does not compose a customer URL from an infrastructure endpoint. connect() prepares token and transport state. It does not block on local policy sync — call start_policy_sync() only when you want in-process Cedar checks.

Call a tool

client.gateway.tools.call() runs an MCP tool through the gateway. Authorization, credential resolution, and audit happen server-side.

result = client.gateway.tools.call(
    "github/list_issues",
    {"repo": "dome"},
    act_as=dome.PlainActAs(email="alice@corp.com"),
)
print(result.structured_content or result.content)

result.content is the raw tuple of MCP content blocks. By default, an upstream isError=true result raises DomeToolExecutionError. Pass raise_on_tool_error=False when you need the failed ToolCallResult and its partial content.

Call an LLM

client.gateway.llm.chat() posts OpenAI-shaped chat requests through the gateway. The response is the provider-shaped dict.

response = client.gateway.llm.chat(
    model="prod-gpt",
    messages=[{"role": "user", "content": "Summarize the open incidents"}],
    act_as=dome.PlainActAs(email="alice@corp.com"),
)

For Anthropic shapes, use client.gateway.llm.messages(model=..., messages=..., max_tokens=...). For provider-native clients, refer to Reference.

Correlate work with an activity

Wrap a run in client.activity(...) to stamp every gateway call and audit event with the same opaque activity ID.

with client.activity(metadata={"case": "incident-123"}) as activity:
    client.gateway.tools.call("github/list_issues", {"repo": "dome"})
    print("activity_id:", activity.activity_id)

Direct calls outside an activity carry no activity ID. The SDK never mints one implicitly.

Shut down

Always close the client to release transports.

client.close()

Or use the context manager form:

with dome.Client(token=..., gateway_url=...) as client:
    client.gateway.tools.call("github/list_issues", {"repo": "dome"})

Next steps

Reference

Full client configuration, act-as methods, errors, and audit reads.

Tutorial

Wire the SDK into a service and verify in audit.

LangChain adapter

Use gateway tools and governed chat models inside LangChain agents.

Data plane concepts

How the gateway authorizes, resolves credentials, and audits every call.

​Install

​Initialize

​Call a tool

​Call an LLM

​Correlate work with an activity

​Shut down

​Next steps

Reference

Tutorial

LangChain adapter

Data plane concepts

Install

Initialize

Call a tool

Call an LLM

Correlate work with an activity

Shut down

Next steps