ElevenLabs MCP and Dome
Voice agents talk to your customers, and you decide what they look up. ElevenLabs Agents reach outside tools through custom MCP servers, added by URL and secret token. Point one at a Dome Gateway, and the voice agent reaches your tools as a registered agent with its own key. Rules decide each call it makes mid-conversation, and audit records every one.
Dome provides a governed gatewayfor ElevenLabs Agents.
A key per voice agent
Each ElevenLabs agent connects with its own Dome key as the secret token. Revoke one key and only that agent loses access.
Reads open, writes closed
Rules decide every tool call the voice agent makes mid-conversation. It looks up a caller's record and can't change it.
Every lookup audited
Each tool call lands in audit under the voice agent's name. You see what it read during every call.
ElevenLabs Agents on a Gatewayin four steps.
Register the voice agent, give it a key, apply the rules, and add the Gateway to ElevenLabs as a custom MCP server. Your tools stay behind the Gateway the whole time.
Register the voice agent
Name the tools it may call and the Gateway it reaches.
shelldome agents register --name support-voice \ --tool hubspot/search_crm_objects \ --tool hubspot/get_crm_objects \ --tool hubspot/search_owners \ --gateway support-gatewayCreate its key
The response carries the key and the Gateway's MCP URL. Store the key in ElevenLabs and nowhere else.
shelldome agents create-key support-voice --name elevenlabs --gateway support-gatewayApply the rules
The Gateway grant opens every tool on the Gateway, and the forbid narrows it to three. Simulate before you deploy.
shelldome rules apply support-voice.cedar \ --agent support-voice --name support-voiceAdd the Gateway to ElevenLabs
In ElevenLabs, open the MCP server integrations and add a custom MCP server. Paste the MCP URL, set the secret token to Bearer and the key, then attach the server to your agent.
Three HubSpot reads,nothing else.
The permit opens HubSpot to the voice agent. The forbid refuses every tool call that isn't one of three reads, so writes and every other tool on the Gateway stay closed.
cedar
permit (principal, action == Dome::Action::"mcp:call", resource is Dome::MCPTool)when { resource.connection_name == "hubspot" };
forbid (principal, action == Dome::Action::"mcp:call", resource is Dome::MCPTool)unless { resource.connection_name == "hubspot" && ["search_crm_objects", "get_crm_objects", "search_owners"].contains(resource.tool_name)};One call,two outcomes.
Switch the caller or the argument and watch the same call decide differently, with every decision landing in audit.
Call
agent support-voice · scheduled
hubspot/search_crm_objects(the contact whose phone is +1 415 555 0123)
- Agentsupport-voice is registered and active
- KeyThe ElevenLabs key belongs to support-voice
- Rulesearch_crm_objects is one of three reads
DecisionAllowed
Three agentsaround ElevenLabs.
Two ElevenLabs voice agents, each with its own key and rules. One of your own agents works from what they recorded.
support-voice
Answer support calls
Looks up the caller's contact record and its owner while it talks, reading HubSpot without changing anything.
billing-voice
Answer billing questions
Reads a caller's invoices in a Stripe sandbox while it talks. Its own key and rules keep it out of HubSpot.
call-summarizer
Log what each call covered
Reads the day's support records and drafts follow-ups for the team lead without ever speaking to a caller.
Bringing it alltogether.
Connecting ElevenLabs Agents to registered tools, models, and identity in Dome completes a governed agent application.
Acting for
Control point
GatewayEvery call decided and audited
RulesGuardsQuotas
Tools
Models
Common questionsabout ElevenLabs Agents.
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.
Runtime
LangGraph
LangGraph agents with their model calls 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.