Voice AI capabilities for your coding agent — powered by ElevenLabs. Works with Cursor, Claude Code, and OpenAI Codex.
Skills live in three trees:
- skills/general/ — general ElevenLabs product skills, copied from the official elevenlabs/skills repository. Don't edit these here; contribute upstream instead.
- skills/mcp/ — plugin-native skills that teach agents how to drive the bundled ElevenLabs MCP server.
- architect/skills/ — Architect skills: opinionated workflows for building and iterating on ElevenLabs agents, mirrored from the internal Architect plugin. Shared payload examples live in architect/reference/.
| Skill | Description |
|---|---|
| agents | Build real-time voice AI agents and assistants |
| dubbing | Dub audio and video into other languages, preserving the original speakers' voices |
| music | Generate instrumental tracks, songs with lyrics, background music, and jingles |
| setup-api-key | Guide users through configuring an ElevenLabs API key |
| sound-effects | Generate sound effects and audio textures from text descriptions |
| speech-engine | Add real-time voice conversations to your own agent runtime |
| speech-to-text | Transcribe audio to text with Scribe v2 |
| text-to-speech | Convert text to natural speech in 70+ languages |
| voice-changer | Transform a recording into a different voice, preserving emotion and timing |
| voice-isolator | Remove background noise and isolate speech from audio |
MCP skills:
| Skill | Description |
|---|---|
| agents-platform | Create, configure, test, and deploy voice agents via the bundled MCP server |
| creative-studio | Generate speech, images, and video, and transcribe audio via the bundled MCP server |
Architect skills:
| Skill | Description |
|---|---|
| agent-simplification | Simplify a complex workflow agent into a leaner architecture, proven equivalent by a test suite |
| architect-agent-memory | Persist facts an agent should remember, in a durable branch-scoped way |
| architect-branches-versions-merge | Work with agent branches, versions, drafts, traffic splits, and merges |
| architect-create-client-tool | Add a client tool that runs in the caller's browser, app, or SDK |
| architect-create-llm-test | Create a single-turn test that checks what the agent says |
| architect-create-simulation-test | Create a multi-turn test where a simulated user talks to the agent |
| architect-create-tool-test | Create a test asserting the agent calls, or avoids, a specific tool |
| architect-create-webhook-tool | Give an agent a new webhook tool that calls an external API |
| architect-edit-existing-tool | Change an existing tool on an agent safely |
| architect-edit-string-fields | Edit long text fields like system prompts without clobbering the rest |
| architect-edit-workflows | Build or change an agent's workflow graph |
| architect-explain-test-runs | Explain why a test or test run passed or failed |
| architect-explore-agent | Map an agent's current setup before changing it |
| architect-generate-agent | Generate a new agent from a description |
| architect-get-agent-ci-green | Drive a red agent test suite to green |
| architect-llm-selection | Choose the right LLM for an agent |
| architect-manage-procedures | Add, edit, rename, and inspect an agent's procedures |
| architect-migrate-from-competitor | Migrate a voice agent from Retell, Vapi, or Bland |
| architect-migrate-to-procedures | Move a system-prompt-driven agent to a procedure-driven setup |
| architect-mock-all-tools | Mock every tool call so simulation tests are deterministic |
| architect-post-call-data | Configure post-call data collection and evaluation criteria |
| architect-review-live-calls | Review recent production conversations and route the findings |
| architect-schedule-launch | Stage a change on a branch and hold the merge until go-live |
| architect-secure-for-production | Make an agent safe, secure, and production-ready |
| architect-structured-procedures | Author structured, deterministic procedures |
| architect-troubleshoot-tool-errors | Diagnose a misbehaving tool from its exact error |
| architect-turn-taking-latency | Tune turn taking and agent response latency |
| architect-update-config-safely | Change agent settings with safe partial patches |
| code-tools | Generate JavaScript/TypeScript modules for ElevenLabs code tools |
| fix-agent-qa-ticket | Fix a QA triage ticket end to end with a verified, branch-scoped fix |
Architect also ships one command: /thermo-nuclear-agent-triage-burndown, which burns down an agent's whole conversation-triage queue — cluster tickets by root cause, fix each cluster, then comment on and resolve every ticket.
The ElevenLabs hosted MCP server — no local install, no API key. Your agent authenticates with your ElevenLabs account over OAuth and gets access to agent management (create, update, list, duplicate, delete agents; inspect widget config, links, and knowledge base size; estimate LLM usage) and text-to-speech generation. Configured in mcp.json.
Install from the Cursor marketplace, or add this repo as a plugin. When Cursor connects to the MCP server, sign in with your ElevenLabs account to complete the OAuth flow.
/plugin marketplace add elevenlabs/plugin
/plugin install elevenlabs@elevenlabs
Or install directly for development: claude --plugin-dir /path/to/plugin
codex plugin marketplace add elevenlabs/plugin
Then install via /plugins in Codex, or invoke skills with $skill-name.
npx skills add elevenlabs/skills- An ElevenLabs account — you'll sign in via OAuth when your agent connects to the MCP server
- Internet access