ElevenLabs integration
In brief
Recommended path today: Server Tool (webhook) → REST. Remote MCP does not yet support Streamable HTTP/SSE required by ElevenLabs.
Prerequisites
- ElevenLabs account with Agents
- AskMarcel API key (ak_live_)
- Voice agent configured in ElevenLabs
Why REST and not MCP?
ElevenLabs requires an MCP server with SSE or Streamable HTTP. Our MCP endpoint returns 405 on GET + Accept: text/event-stream. Webhook tool → REST is guaranteed and tested.
MCP in preparation
Streamable HTTP support is planned. This page will be updated when SSE transport ships.
Configure a Server Tool
- 1
Open your ElevenLabs agent
Agents → your agent → Tools → Add tool → Webhook.
- 2
URL and method
POST https://api.askmarcel.app/v1/diagnostic
- 3
Headers
Authorization: Bearer YOUR_AK_LIVE_KEY and Content-Type: application/json.
- 4
Parameters
Define brand (string), model (string), error_code (string) as parameters the agent can extract from conversation.
Body template{ "brand": "{{brand}}", "model": "{{model}}", "error_code": "{{error_code}}" } - 5
System prompt
Instruct the agent to always cite the source (manual, page) returned by the API before giving a procedure.
Handle the response
Map steps[] for voice readout and source.page for traceability. confidence.level = low → ask technician for confirmation.
Verify it works
Test the agent with "I have error E7 on a Mitsubishi Ecodan". The agent should call the tool and read the steps.
curl -s https://api.askmarcel.app/v1/diagnostic \
-H "Authorization: Bearer ak_live_..." \
-H "Content-Type: application/json" \
-d '{"brand":"Mitsubishi","model":"Ecodan PUHZ-W85","error_code":"E7"}'Known limitations
- Native ElevenLabs MCP not supported today (no SSE)
- Voice latency: plan a hold message during API call (~200-500ms warm)
- Developer tier not allowed in production
Video coming soon
Notify me