Connect your agent

MCP

Use your Timbre deployment's /mcp URL and an API key from Settings. Send the key as Authorization: Bearer <key>. The server also supports OAuth for clients that discover and request access through the sign-in flow.

Tools: whoami, echo, list_voices, list_channels, get_tone, record_edit, record_outcome, add_sample, propose_rule, propose_rule_change.

Read get_tone(voice, channel) before drafting. Record the original draft and human final with record_edit. Send only actual post numbers to record_outcome. Unknown voice slugs are created on writes.

Voice skill

The repository's skills/voice/SKILL.md provides /voice, /voice learn, /voice add, /voice show, and named brand voices. It keeps no local tone files. Rules are proposed by the server and approved on your voice page.

SessionStart hook

Configure TIMBRE_URL, TIMBRE_API_KEY and optionally TIMBRE_DEFAULT_VOICE (default hauke) in ~/.config/timbre/env. Register hooks/timbre-tone.sh as a SessionStart command. The hook fetches voice-wide rules and stays silent when unconfigured or unreachable. Resumes load tone only when the approved profile has changed.

For local development use port 8081 and the shared Postgres/Mailpit services. See the repository README for exact setup and verification commands.