Announcing Dome Managed Agents.

Try it now

Claude Agent SDK and Dome

The Claude Code loop, with every call governed. Point a Claude Agent SDK agent at the Gateway for its model and its MCP tools. The agent loop is the one Claude Code runs. Every model call and tool call is checked against your rules and recorded against the agent.

Dome governs every tool and model callin Claude Agent SDK.

Only the tools you grant

Switch off the built-in tools and the agent works through the Gateway alone. It sees only the tools its agent is granted.

Claude through the Broker

The SDK's Claude Code CLI takes the Gateway as its Anthropic base URL. The model name is a pool, with failover across providers.

One key, one record

The agent's Dome key authenticates both the model and the tools. Every call lands in one audit trail against that agent.

Claude Agent SDK on Domein two changes.

Give the SDK the Gateway's address for models and add the Gateway as an HTTP MCP server, leaving the agent loop untouched.
  1. Install the SDK

    Anthropic's Agent SDK for Python bundles the Claude Code CLI it drives and needs Python 3.10 or later.

    shell
    pip install claude-agent-sdk
  2. Load tools from the Gateway

    The Gateway's MCP endpoint serves the tools this agent is granted. An empty tools list switches off the built-in file and shell tools.

    python
    mcp_servers={"dome": {    "type": "http",    "url": f"{GATEWAY_URL}/mcp",    "headers": {"Authorization": f"Bearer {DOME_AGENT_KEY}"},}},strict_mcp_config=True,tools=[],allowed_tools=["mcp__dome"],
  3. Call models through the Broker

    The CLI reads its Anthropic base URL and bearer token from the environment, and the model name is a pool.

    python
    model="claude-opus-5-5",env={"ANTHROPIC_BASE_URL": GATEWAY_URL, "ANTHROPIC_AUTH_TOKEN": DOME_AGENT_KEY},

One agent,end to end.

The Claude Code loop as a library: query() runs Opus through the Broker with the Gateway as its MCP server. Its Dome key is the only credential it holds.

researcher

Answer questions about the code

query() streams messages until the result arrives. The Gateway decides each tool call, and the Broker routes each model call to a provider that answers.

python
import asyncio
from claude_agent_sdk import ClaudeAgentOptions, ResultMessage, query
options = ClaudeAgentOptions(    # Model: a pool on the Broker, through its Anthropic-compatible endpoint    model="claude-opus-5-5",    env={"ANTHROPIC_BASE_URL": GATEWAY_URL, "ANTHROPIC_AUTH_TOKEN": DOME_AGENT_KEY},    # Tools: whatever this agent is granted on the Gateway, and nothing local    mcp_servers={"dome": {        "type": "http",        "url": f"{GATEWAY_URL}/mcp",        "headers": {"Authorization": f"Bearer {DOME_AGENT_KEY}"},    }},    strict_mcp_config=True,    tools=[],    allowed_tools=["mcp__dome"],    system_prompt="Answer questions about the code. Search before you answer.",)

async def main():    async for message in query(prompt="Where is rate limiting handled?", options=options):        if isinstance(message, ResultMessage):            print(message.result)

asyncio.run(main())

Bringing it alltogether.

Connecting Claude Agent SDK agents to tools, models, and identity in Dome completes a governed agent application.
Dome

Agent

Claude Agent SDKRuntime
Claude CodeClient

Control point

GatewayEvery call decided and audited
RulesGuardsQuotas

Common questionsabout Claude Agent SDK.

Can the Claude Agent SDK use a custom Anthropic base URL?
Yes. Set ANTHROPIC_BASE_URL to the Gateway and ANTHROPIC_AUTH_TOKEN to the agent's Dome key in the SDK's env option.
How do I connect the Claude Agent SDK to a remote MCP server?
Add it under mcp_servers with type http, the Gateway's /mcp URL, and the Dome key as a bearer token. Allow it with mcp__dome in allowed_tools.
Can the agent still run shell commands or edit files?
Not with tools set to an empty list. The agent then works only through the tools the Gateway grants it.
Where are rules enforced?
At the Gateway, on every model call and every tool call. A refused model call ends the run with the reason.

More of whatDome works with.

Model

Claude Fable

Fable 5.1 from Anthropic and Amazon Bedrock in one failover pool, open to one group and capped by quota.
Provider

Anthropic

The Claude API behind the Model Broker: the key held in Dome, every call authorized, metered and audited.
MCP server

Atlassian

The Atlassian Rovo MCP server behind the Tool Gateway, with rules that decide each Jira and Confluence call on its tool and site.
Identity

Okta

Okta tokens verified on every agent call, so rules and audit name the person each agent acted for.
Client

Codex

Codex on a Dome Gateway with per-developer sign-in, rules on every tool call, and audit by name.
Agent service

TinyFish

TinyFish's web agents behind the Tool Gateway, with rules that decide each run on the site it targets.

Talk with ourFDE team.

Or start free, and register your first agent in minutes. Connect it through a gateway, give it rules, and see every call it makes in one audit trail.

Talk with our FDE team

Our engineers design, build and run governed agents for the part of your business you want to change. You see what they deliver.

Agent frameworks and runtimes: govern agents without a rewrite

Build agents on the framework you already use. Dome governs their tool and model calls without a rewrite.

Every agent you run, under one system of control.