generated: '2026-09-17'
method: probed
source: |
  https://api.elevenlabs.io/.well-known/oauth-protected-resource (RFC 9728, HTTP 200),
  POST https://api.us.elevenlabs.io/v1/mcp tools/list (HTTP 401 + WWW-Authenticate Bearer
  resource_metadata challenge), https://elevenlabs.io/docs/eleven-agents/operate/hosted-mcp,
  https://elevenlabs.io/llms.txt, and https://github.com/elevenlabs/elevenlabs-mcp.
status: gated
server:
  name: elevenlabs
  transport: http
  url: https://api.elevenlabs.io/v1/mcp
  vendor: ElevenLabs
  first_party: true
deployment:
  mode: both
  endpoint: https://api.elevenlabs.io/v1/mcp
  install: uvx elevenlabs-mcp
  package: https://pypi.org/project/elevenlabs-mcp/
  auth: oauth
  verified: probed
  checked: '2026-09-17'
  note: |
    BOTH shapes ship, and they are different products. The hosted REMOTE server at
    https://api.elevenlabs.io/v1/mcp is callable by an MCP client today over OAuth with
    nothing installed — Anthropic lists it in the Claude Desktop connector directory. The
    open-source LOCAL stdio server (pypi elevenlabs-mcp 0.12.2,
    github.com/elevenlabs/elevenlabs-mcp) is a separate package a human installs and runs.
    The hosted server covers ElevenAgents management plus Text to Speech; the local server
    covers the generation surface.
endpoints:
- region: global
  url: https://api.elevenlabs.io/v1/mcp
  source: https://elevenlabs.io/docs/eleven-agents/operate/hosted-mcp
- region: us
  url: https://api.us.elevenlabs.io/v1/mcp
  source: https://api.elevenlabs.io/.well-known/oauth-protected-resource
  note: the only endpoint named by a machine-readable discovery document
- region: eu
  url: https://api.eu.residency.elevenlabs.io/v1/mcp
  source: https://elevenlabs.io/docs/eleven-agents/operate/hosted-mcp
- region: in
  url: https://api.in.residency.elevenlabs.io/v1/mcp
  source: https://elevenlabs.io/docs/eleven-agents/operate/hosted-mcp
- region: sg
  url: https://api.sg.residency.elevenlabs.io/v1/mcp
  source: https://elevenlabs.io/docs/eleven-agents/operate/hosted-mcp
authorization:
  scheme: oauth2
  protected_resource_metadata: https://api.elevenlabs.io/.well-known/oauth-protected-resource
  authorization_server_metadata: https://api.elevenlabs.io/.well-known/oauth-authorization-server
  issuer: https://api.us.elevenlabs.io
  pkce: S256
  bearer_methods_supported:
  - header
  scopes_supported:
  - convai_read
  - convai_write
  - text_to_speech
  - speech_history_read
  - flows
  - image_video_generation
  - voice_generation
  dynamic_client_registration: false
  client_id_metadata_document_supported: true
  note: See scopes/elevenlabs-scopes.yml. No DCR endpoint is advertised; the server relies
    on CIMD (client_id metadata documents) so hosted clients need no registration.
probe:
  url: https://api.us.elevenlabs.io/v1/mcp
  method: POST
  body: '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
  http_status: 401
  www_authenticate: Bearer resource_metadata="https://api.us.elevenlabs.io/.well-known/oauth-protected-resource"
  response: '{"detail":"OAuth bearer token required for the hosted MCP."}'
  verdict: reachable, first-party, auth-gated. The RFC 9728 challenge is itself the
    strongest evidence this is a real MCP server rather than a guessed path.
tools_note: |
  No machine-readable tool list captured — anonymous tools/list is OAuth-gated. The
  capability list below is the PROVIDER'S OWN prose enumeration from the hosted-MCP doc,
  not an introspected manifest: names and inputSchemas require an authenticated tools/list.
  Do not treat capabilities[] as tool identifiers.
capabilities_documented:
- Create new agents from a natural-language description
- Update any agent setting (system prompt, voice, language, first message)
- List agents and inspect or compare their configurations
- Review an agent's recent conversations and read full transcripts
- Explore the topics an agent's conversations cover
- Duplicate and delete agents
- Estimate an agent's expected LLM usage and cost before making changes
- Retrieve an agent's widget configuration and shareable link
- Check the size of an agent's knowledge base
- Generate speech audio from text, returned as a short-lived download link
capabilities_source: https://elevenlabs.io/docs/eleven-agents/operate/hosted-mcp
safety_notes:
- Deleting an agent is destructive; the provider explicitly tells operators to review tool
  calls before approving them and to restrict write access.
- Workspace administrators can disable individual tools for the whole organization; a tool
  an admin disables cannot be re-enabled by a user.
local_server:
  name: elevenlabs-mcp
  repository: https://github.com/elevenlabs/elevenlabs-mcp
  package: https://pypi.org/project/elevenlabs-mcp/
  version: 0.12.2
  transport: stdio
  auth: api-key
  auth_note: the local server reads ELEVENLABS_API_KEY; the hosted server uses OAuth and
    stores no key in the client.