Announcing Dome Managed Agents.

Try it now

Retell AI MCP server and Dome

Agents that place phone calls, on numbers you allow. Retell's MCP server gives an agent the whole Retell API through one tool. Connect it to Dome once, and rules decide each call on the endpoint it names and the number it dials. Every request lands in one audit trail.

Dome provides a governed gatewayfor Retell AI.

Endpoints on a list

Retell's MCP server reaches every endpoint through invoke_api_endpoint. Rules read the endpoint name, so batch calls and agent edits stay closed.

+1 numbers only

Every createPhoneCall carries a to_number. The rule refuses any that isn't a +1 number.

One voice agent, no overrides

Rules pin the Retell agent that speaks on each call. A call that rewrites the agent's configuration is refused.

Retell behind the Gatewayin four steps.

Add Retell's server, sync its three tools, deploy the endpoint list and cap each agent's calls.
  1. Add the Retell MCP server

    Dome holds the Retell API key and sends it as a Bearer token, out of reach of your agents.

    shell
    dome tools add --name retell \  --url https://mcp.retellai.com \  --auth-method api-key \  --credential-type shared \  --authorization "Bearer $RETELL_API_KEY" \  --gateway voice-gateway
  2. Sync the catalog

    Sync lists Retell's three tools: list_api_endpoints, get_api_endpoint_schema, and invoke_api_endpoint.

    shell
    dome tools catalog sync retell
  3. Apply the rules

    The endpoint list is scoped to appointment-confirmer. Simulate a batch call to see it refused, then deploy.

    shell
    dome rules apply retell-calls.cedar \  --agent appointment-confirmer --name retell-calls
  4. Cap the agent's calls

    The appointment-confirmer agent gets a hundred calls a day, and every tool call counts, reads included.

    shell
    dome quotas set --subject agent --agent appointment-confirmer \  --dimension tool --unit calls --limit 100 --window daily

Read calls,place them on +1 numbers.

The permit opens Retell to the agent. The forbid refuses any endpoint but four reads and createPhoneCall, and refuses a call unless it dials a +1 number through one Retell agent with no override.
cedar
permit (principal, action == Dome::Action::"mcp:call", resource is Dome::MCPTool)when { resource.connection_name == "retell" };
forbid (principal, action == Dome::Action::"mcp:call", resource is Dome::MCPTool)when {  resource.connection_name == "retell" &&  resource.tool_name == "invoke_api_endpoint"}unless {  resource has arguments &&  resource.arguments has endpoint &&  (["listCalls", "getCall", "listAgents", "getAgent"].contains(resource.arguments.endpoint) ||   (resource.arguments.endpoint == "createPhoneCall" &&    resource.arguments has arguments &&    resource.arguments.arguments has requestBody &&    resource.arguments.arguments.requestBody has to_number &&    resource.arguments.arguments.requestBody has override_agent_id &&    resource.arguments.arguments.requestBody.to_number like "+1*" &&    resource.arguments.arguments.requestBody.override_agent_id == "agent_appointment_confirm" &&    !(resource.arguments.arguments.requestBody has agent_override)))};

One call,two outcomes.

Switch the caller or the argument and watch the same call decide differently, with every decision landing in audit.

Number

agent appointment-confirmer · scheduled
retell/invoke_api_endpoint(endpoint: "createPhoneCall", to_number: "+14155550123", override_agent_id: "agent_appointment_confirm")
  1. Agentappointment-confirmer is registered and active
  2. EndpointcreatePhoneCall is on the list
  3. Rule+1 numbers through agent_appointment_confirm
DecisionAllowed

Three agentson Retell.

One places calls, one reads them, and one builds. Each gets its own rules on the same three tools.

appointment-confirmer

Call patients about tomorrow

Calls each patient booked for the next day through one Retell agent, and is refused for any number outside +1.

call-analyst

Summarize the week's calls

Lists and reads calls, then reports outcomes to the clinic manager, using read endpoints only.

intake-builder

Draft new voice agents

Creates draft Retell agents for an engineer to test. It can't publish a version or place a call.

Bringing it alltogether.

Connecting Retell AI to registered agents, tools, and models in Dome completes a governed agent application.
Dome

Control point

GatewayEvery call decided and audited
RulesGuardsQuotas

Tools

Retell AIAgent service
SalesforceMCP server

Common questionsabout Retell AI.

Does Retell AI have an MCP server?
Yes, at mcp.retellai.com. Dome connects by URL with your Retell API key.
How do rules work when one tool reaches the whole Retell API?
Rules read the endpoint argument of invoke_api_endpoint. Allow the endpoints an agent needs and the rest are refused.
Can a Retell agent dial international numbers?
Not under this rule. It reads to_number on each createPhoneCall and refuses anything outside +1.
Can an agent change what the Retell agent says on a call?
Not with this rule. It pins override_agent_id and refuses any call that carries an agent_override.

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

Okta

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

OpenAI Agents SDK

OpenAI Agents SDK agents with their models on the Model Broker and their MCP tools on the Tool Gateway.
Client

Claude Code

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

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 services: web, research, payments, voice and sandboxes

Web actions, research, payments, phone calls and code execution are where agents reach past your walls. Put each service behind the Gateway, and every call is authorized, metered and audited.

Every agent you run, under one system of control.