Reusable Agent Skills for building with the Speechify voice API — text-to-speech, streaming, speech marks, voice cloning, SSML, multilingual synthesis, and using Speechify as the TTS leg inside voice agents.
They work with any Agent Skills–compatible coding assistant (Claude Code, Cursor, Codex, and ~75 others) and follow one rule above all: never invent Speechify API facts — verify them against the live source.
/plugin marketplace add Speechify-AI/skills
/plugin install speechify@speechify-skills
This installs all the skills and auto-registers the hosted ask-speechify
MCP (https://mcp.speechify.ai/mcp, public, no key) so skills can verify facts
against the live source — no separate setup step. Skills are then invokable as
/speechify:<skill-name>.
Vercel's open skills CLI installs the
raw skill folders into Claude Code, Cursor, Codex, and ~75 others:
npx skills add Speechify-AI/skills # all skills
npx skills add Speechify-AI/skills --skill speechify-text-to-speech # just one
npx skills add Speechify-AI/skills -g # install globally
npx skills add Speechify-AI/skills -a claude-code -a cursor # pick agentsInstalling this way doesn't bundle the MCP — run the speechify-setup-mcp skill
(or speechify mcp install) to connect it. Every skill offers to do this if the
MCP isn't present.
| Skill | Use it to… |
|---|---|
speechify |
Entry point — routes to the right skill and carries the core doctrine. |
speechify-setup-api-key |
Get, validate, and configure SPEECHIFY_API_KEY. |
speechify-text-to-speech |
Synthesise speech from text or SSML (/v1/audio/speech). |
speechify-streaming |
Low-latency streaming (/v1/audio/stream and /stream/with-timestamps). |
speechify-speech-marks |
Word/char timestamps → captions, subtitles, highlighting. |
speechify-voices |
List, filter, and choose voices. |
speechify-voice-cloning |
Clone a voice from a sample and synthesise with it. |
speechify-ssml |
SSML structure and emotion/prosody control. |
speechify-multilingual |
Multilingual synthesis; language/locale/accent. |
speechify-audio-formats |
Pick the right output format (web, telephony, PCM). |
speechify-models |
Model selection + deprecation/version discipline. |
speechify-voice-agents |
Speechify as the TTS leg in LiveKit / Pipecat / Vapi / Twilio / AI SDK. |
speechify-sdks |
Install/init the Python & TypeScript SDKs; native vs SDK. |
speechify-cli |
Install and use the speechify CLI (@speechify/cli) from a terminal/CI. |
speechify-setup-mcp |
Connect the hosted Speechify knowledge MCP (mcp.speechify.ai) to your editor. |
speechify-cookbook |
Match a task to one of the cookbook recipes. |
speechify-demos |
Match a use case to a clonable demo. |
Model IDs, voice IDs, endpoints, request fields, output formats, and version dates change. Every skill defers those to a live source instead of baking them in:
ask-speechifyMCP — grounded answers over Speechify's repos, SDKs, API reference, and docs. Hosted athttps://mcp.speechify.ai/mcp(public, no key); connect it with thespeechify-setup-mcpskill. Every skill offers to set it up if it isn't connected.- Docs —
https://docs.speechify.ai(append.mdto any page for agent-readable Markdown). - SDKs —
models.list()/voices.list()return the current catalogue.
If a fact isn't confirmed by one of those, the skill tells the agent to go check rather than guess.
The set is a superset distilled from the leading voice/AI-infra skill repos
(Cartesia, Deepgram, ElevenLabs, LiveKit, Pipecat, OpenAI/Codex, Vercel), mapped
onto Speechify's product surface. The full rationale — the competitor matrix,
the conventions adopted, and what was deliberately deferred — is in
docs/alignment.md.
Authoring conventions and a skill template are in
CONTRIBUTING.md and templates/SKILL.template.md.
MIT — see LICENSE.