Zed (1.0, released April 2026) supports MCP via its context_servers config.
pip install --user claude-code-talkerOpen Zed → Settings (cmd-,) → settings.json. Add a context_servers block:
{
"context_servers": {
"codetalker": {
"command": {
"path": "codetalker-mcp",
"args": [],
"env": {}
}
}
}
}If codetalker-mcp isn't on Zed's PATH, use the absolute path:
"path": "/Users/you/.local/bin/codetalker-mcp"Zed's Agent Panel will discover the new context server on startup.
Zed deliberately requires explicit permission for each tool call (its "permissioned per-action" model). The first time the Agent Panel wants to call mcp__codetalker__tts_speak, you'll see a permission prompt — accept "allow always" for the tools you trust.
Recommended per-action allowlist:
tts_status— always allowtts_set_mode— always allowtts_mute/tts_unmute— always allowtts_speak— ask each time (prevents surprise narration)tts_shutdown— ask each time (rare)
In Zed's Agent Panel:
Use codetalker.tts_status and tell me the state.
You should see a permission prompt followed by the status output.
- Zed 1.0's parallel agents feature is a perfect match for codetalker — when multiple agents run in parallel, codetalker's per-session mode + cadence settings (via the webui dashboard) let you narrate one of them in
livemode while the others arebriefor muted. Set this up viaclaude-code-talker serve+ open the dashboard at http://localhost:17832. - If you use Zed's collaboration feature (shared editing), only the host's codetalker daemon narrates — codetalker is local-first by design.