Announcing Dome Managed Agents.

Try it now

CrewAI agents and Dome

Keep your crew, and govern every call it makes. A crew is several agents, and Dome can tell them apart. Give each one an LLM pointed at the Broker, the Gateway as its MCP server, and its own Dome key. Roles, tasks and process stay as written, and rules decide each member's calls.

Dome governs every tool and model callin CrewAI.

Roles and tasks stay as they are

Agents, tasks, and process are unchanged, and only each agent's LLM and MCP server change.

A key per agent

Give each crew member its own Dome key and rules can tell them apart. The researcher can read what the writer cannot.

Who in the crew did it

Audit records each LLM and tool call against the crew member that made it, and the person it acted for.

CrewAI on Domein two changes.

Each agent gets a crewai.LLM on the Broker and an MCPServerHTTP on the Gateway, and the crew definition is untouched.
  1. Install CrewAI

    CrewAI includes its MCP client and runs on Python 3.10 to 3.13.

    shell
    pip install crewai
  2. Load tools from the Gateway

    Each agent's mcps list takes the Gateway, and sees the tools that agent's key is granted.

    python
    from crewai.mcp import MCPServerHTTP
    dome = MCPServerHTTP(    url=f"{GATEWAY_URL}/mcp",    headers={"Authorization": f"Bearer {DOME_AGENT_KEY}"},)
  3. Call models through the Broker

    Use an OpenAI-compatible LLM, set the Gateway as its base URL, and name a pool as the model.

    python
    from crewai import LLM
    llm = LLM(    model="openai/claude-opus-5-5",    base_url=f"{GATEWAY_URL}/v1",    api_key=DOME_AGENT_KEY,)

One agent,end to end.

A one-agent crew: the LLM points at the Broker and MCPServerHTTP points at the Gateway. Add crew members and each gets its own Dome key.

researcher

Answer questions about the code

The crew's one task is the question. kickoff() runs it through the GitHub tools, each call decided at the Gateway, each model call routed by the Broker.

python
from crewai import LLM, Agent, Crew, Taskfrom crewai.mcp import MCPServerHTTP
# Model: a pool on the Broker, through the OpenAI-compatible endpointllm = LLM(    model="openai/claude-opus-5-5",    base_url=f"{GATEWAY_URL}/v1",    api_key=DOME_AGENT_KEY,)
# Tools: whatever this agent is granted on the Gatewaydome = MCPServerHTTP(    url=f"{GATEWAY_URL}/mcp",    headers={"Authorization": f"Bearer {DOME_AGENT_KEY}"},)
researcher = Agent(    role="researcher",    goal="Answer questions about the code",    backstory="Searches repositories before answering.",    llm=llm,    mcps=[dome],)
task = Task(    description="Where is rate limiting handled?",    expected_output="The files and functions that handle rate limiting.",    agent=researcher,)
result = Crew(agents=[researcher], tasks=[task]).kickoff()print(result.raw)

Bringing it alltogether.

Connecting CrewAI crews to tools, models, and identity in Dome completes a governed agent application.
Dome

Acting for

Agent

CrewAIRuntime

Control point

GatewayEvery call decided and audited
RulesGuardsQuotas

Common questionsabout CrewAI.

How do I use a custom LLM base URL in CrewAI?
Pass base_url and api_key to crewai.LLM, with the model as openai/ plus the pool name, since Dome's /v1 endpoint is OpenAI-compatible.
How do I connect CrewAI to a remote MCP server?
Give the agent an MCPServerHTTP in its mcps list, with the Gateway's /mcp URL and the Dome key as a bearer token.
Can each agent in a crew have different permissions?
Yes. Register each agent in Dome with its own key, and rules decide per agent.
Do rules run inside the crew?
No. They run at the Gateway, on every LLM and tool call. A refused call comes back as a 403 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

GitHub

The GitHub MCP server behind the Tool Gateway, with rules that decide each call on its owner and repository.
Identity

Microsoft Entra ID

Entra access tokens verified on every agent call, so rules read app roles and audit names the person.
Client

Claude Code

Claude Code 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.