v0.1.0-beta.7

Directory readiness, no behavior change. Every tool now carries its title in the MCP annotations as well as at the top level, and the Desktop extension manifest's author points at the paxalabs GitHub organization. Both are requirements for Anthropic's featured extensions directory.

Everything from beta.6 (cached hook phrases, --cache for custom phrases, faster player start, shorter phrases) is unchanged.

Install

• Claude Code: claude mcp add paxa -e PAXA_API_KEY=pxa_... -- npx -y @paxalabs/mcp
• Claude Desktop: download paxalabs-mcp-0.1.0-beta.7.mcpb below and open it.
• Command line: npm install -g @paxalabs/mcp (hooks run this one; upgrade with the same command)

See "Voice mode for Claude Code" in the README for the hook configuration.

v0.1.0-beta.6

Faster, free hook notifications. paxa say now synthesizes each built-in hook phrase once per voice and keeps the audio in your user cache directory, so later notifications play from disk: no network round trip and no credits. On a Mac a permission notification went from about 2.3 s to 1.4 s, and 1.25 s of that is the voice itself.

Custom phrases. Write your own text in the hook command and add --cache so it gets the same treatment, one hook entry per matcher:

{ "matcher": "permission_prompt", "hooks": [{ "type": "command", "command": "paxa say --cache \"Hey, need your OK\"" }] }

Without --cache, nothing you type or pipe into paxa say is written to disk, and messages carried inside a hook payload never are.

Faster player start. Cached phrases play through the installed player with the smallest start-up cost: mpg123 (about 50 ms), then ffplay or mpv (about 300 ms), falling back to the platform player. On macOS the built-in afplay needs 0.4 to 0.9 s just to start and stop, so brew install mpg123 makes a real difference for short phrases.

Also

• paxa say no longer loads the MCP server, so it starts in about 20 ms instead of 70 to 100.
• Shorter built-in phrases: "Permission needed.", "Waiting for you.", "Question for you.", "Input needed.", "Finished.", "Done.", "Stopped with an error."

Install

• Claude Code: claude mcp add paxa -e PAXA_API_KEY=pxa_... -- npx -y @paxalabs/mcp
• Claude Desktop: download paxalabs-mcp-0.1.0-beta.6.mcpb below and open it.
• Command line: npm install -g @paxalabs/mcp (hooks run this one; upgrade with the same command)

See "Voice mode for Claude Code" in the README for the hook configuration.

v0.1.0-beta.5

Fixes a broken 0.1.0-beta.4. Adding a second command in beta.4 meant a bare npx -y @paxalabs/mcp could not decide which one to run, so the MCP server failed to start for every client using the documented command. Upgrade if you are on beta.4.

One command now. The package ships a single paxa entry point:

Command	What it does
paxa	starts the MCP server (this is what npx runs)
paxa say "Build finished"	speaks a line
paxa say with a Claude Code hook payload on stdin	speaks the matching phrase

This replaces the paxalabs-mcp and paxa-say commands from earlier betas. If you configured a hook with paxa-say, change it to paxa say.

Install

• Claude Code: claude mcp add paxa -e PAXA_API_KEY=pxa_... -- npx -y @paxalabs/mcp
• Claude Desktop: download paxalabs-mcp-0.1.0-beta.5.mcpb below and open it.
• Command line: npm install -g @paxalabs/mcp

See "Voice mode for Claude Code" in the README for the hook configuration.

v0.1.0-beta.4

New in beta.4

• paxa-say: a command shipped in the package that speaks a line from the terminal or turns a Claude Code hook into a spoken phrase ("Claude needs your permission."), so Claude Code can call you when it is waiting for input. See "Voice mode for Claude Code" in the README for the hook config and the CLAUDE.md snippet that makes Claude speak summaries.
• Voice guidance in the tools: English text usually sounds best with an English voice; the default stays nomyen.

Install

• Claude Code: claude mcp add paxa -e PAXA_API_KEY=pxa_... -- npx -y @paxalabs/mcp
• Claude Desktop: download paxalabs-mcp-0.1.0-beta.4.mcpb below and open it with Claude Desktop.
• paxa-say on your PATH: npm install -g @paxalabs/mcp

Beta: tool names and behavior may still change before 1.0. Please report problems in the issues.

v0.1.0-beta.3

First public beta of the Paxa Labs MCP server.

Install

• Claude Code: claude mcp add paxa -e PAXA_API_KEY=pxa_... -- npx -y @paxalabs/mcp
• Claude Desktop: download paxalabs-mcp-0.1.0-beta.3.mcpb below, open it with Claude Desktop, and enter your API key in the extension settings. The bundle carries its own copy of the server, so no Node.js install is needed.
• Any MCP client: npx -y @paxalabs/mcp with PAXA_API_KEY in the environment.

What is in this release

• Ten tools: speak, queue_speech, control_playback, play_audio, text_to_speech, translate_to_thai, ocr_document, list_voices, list_models, get_account. Every paid tool states its price.
• Streaming playback: with ffplay, mpv, or mpg123 installed, speech starts on the first bytes from the API, about 0.3 s to the first word regardless of length.
• One ordered audio queue, so sounds never overlap; long reads synthesize ahead and play gap-free.
• Tool annotations and per-error handling hints, so agents know what to do or what to tell the user when something fails, including a missing API key.
• Listed on the official MCP Registry as com.paxalabs/mcp.

Beta: tool names and behavior may still change before 1.0. Please report problems in the issues.