> ## Documentation Index > Fetch the complete documentation index at: https://docs.tuple.app/llms.txt > Use this file to discover all available pages before exploring further. # MCP server `tuple mcp` runs a [Model Context Protocol](https://modelcontextprotocol.io) server over stdin/stdout. It exposes Tuple's call, contact, screen, transcription, notification, and past-call transcript capabilities to AI coding agents such as Claude Code, Claude Desktop, and OpenAI Codex. The MCP server is a sibling of the [`tuple` CLI](/cli/commands): both require the Tuple desktop app to be running and signed in. ## Install with a coding agent Run `tuple mcp install` and pick an agent, or specify one directly: ```bash theme={null} tuple mcp install # interactive picker tuple mcp install claude # Claude Code CLI tuple mcp install claude-desktop # Claude Desktop app tuple mcp install codex # OpenAI Codex ``` You can install Tuple in multiple agents at once: ```bash theme={null} tuple mcp install claude codex ``` By default the server is registered as `tuple`. Pass `--name` to use a different identifier in your agent config. Each install command writes to that agent's standard config: | Agent | Config file | | ---------------- | ---------------------------------------- | | `claude` | `~/.claude.json` | | `claude-desktop` | Claude Desktop's per-OS config directory | | `codex` | `~/.codex/config.toml` | Restart the agent after installing. ## Tools Once installed, the agent can call these tools. Names mirror the CLI commands. ### State and contacts | Tool | What it does | | ----------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `get_state` | Bounded orientation summary: `in_call`, normalized `call`, user identity, contact and room counts (with a capped sample of online favorites), pending invites, and connection status. A `see` map points at the entity tools (`list_contacts`, `list_rooms`, `get_active_call`, `get_current_user`) for full drill-down. Constant-size regardless of team size. | | `get_current_user` | Identity of the authenticated user (id, email, full name, short name). Use instead of `get_state` when you only need to know who is logged in. | | `list_contacts` | List or search contacts. Optional `query`, `status` (`online`/`busy`/`offline`, with `available` as a synonym for `online`), `kind` (`teammate`/`external`), `favorited`, and `recent` filter params compose with AND. | | `favorite_contact` / `unfavorite_contact` | Pin or unpin a contact. | | `remove_contact` | Remove a contact (destructive; agents prompt to confirm). | | `invite_contact` | Send a Tuple invite by email (mutating; agents prompt to confirm). | ### Calls | Tool | What it does | | ---------------------------------------- | ------------------------------------------------ | | `start_call` | Start a 1:1 call with a contact. | | `join_call` | Join a contact's active call or a room URL/slug. | | `end_call` | Leave the active call. | | `mute` / `unmute` | Mute or unmute your microphone. | | `add_participant` / `remove_participant` | Manage participants in the active call. | | `capture_screen` | Return a JPEG of the currently shared screen. | ### Rooms | Tool | What it does | | ----------------------------------- | --------------------------------------------------------------------------------------------------------- | | `list_rooms` | List personal and team rooms. Supports kind, occupied, favorite, member-detail, and result-limit filters. | | `join_room` | Join a room by slug or URL without resolving the argument as a contact. | | `favorite_room` / `unfavorite_room` | Pin or unpin a room. | | `create_room` | Create a team room. | ### Transcription and notifications | Tool | What it does | | -------------------------------------------- | --------------------------------------------------------------------------------------------------------- | | `start_transcription` / `stop_transcription` | Toggle transcription for a call. | | `get_transcription` | Return transcription segments plus human-readable transcript text. | | `get_transcription_events` | Return structured lifecycle events for a call. | | `notify_user` | Post a fire-and-forget notice (title, body, sender). | | `request_user_decision` | Post a yes/no card (title, body, sender, optional accept/reject labels) and block until the user answers. | | `list_notifications` | Get the history of notification events for a call. | | `cancel_notification` | Withdraw a card the agent posted earlier. | ### Stored transcripts (past calls) These tools query the local transcript database of past transcribed calls. They work without an active call. See [Searching past calls](/pairing-with-tuple/searching-past-calls) for the underlying CLI. | Tool | What it does | | ---------------------- | ---------------------------------------------------------------------------------------------- | | `transcription_search` | Full-text search across past call transcripts, with filters and optional conversation context. | | `transcription_list` | List past recorded calls, newest first, with dates, participants, and transcript sizes. | | `transcription_show` | Return a past call's full transcript across all its recording sessions. | | `transcription_export` | Write past call transcripts to disk as Markdown, text, or NDJSON. | | `delete_stored_call` | Permanently delete a stored call's recordings, transcript, and events. | Destructive and mutating tools (`remove_contact`, `invite_contact`, `end_call`, `delete_stored_call`, etc.) carry MCP annotations that drive the client's confirmation prompt — the user approves them at call time. ## Prompts The server also registers a small set of [MCP prompts](https://modelcontextprotocol.io/specification/server/prompts). MCP-aware agents surface these as slash commands — for example, Claude Code shows them under `/tuple:` once Tuple is installed. | Slash command | Arguments | What it does | | ------------------------ | ----------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `/tuple:connect` | `call_id` (defaults to `current`), `tail` (default `100`) | Drop the agent into a live Tuple call as a silent sidekick. Returns recent lifecycle events and the trailing transcript so the agent can join mid-call without re-reading the whole history, plus instructions for following the unified stream, capturing the shared screen, and knowing when to speak. | | `/tuple:summarize_call` | `call_id` (defaults to `current`) | Pull the current call's full transcript and events and ask the agent for a structured summary: key discussion points, decisions, action items, and open follow-ups. | | `/tuple:find_discussion` | `topic` (required, FTS5 syntax), `participant`, `after`, `before` | Search past recorded calls for a topic and synthesize where, when, and by whom it was discussed, with conversation excerpts around each match. | ### `/tuple:connect` Use `/tuple:connect` from inside an MCP-aware agent (Claude Code, Claude Desktop, Codex, or anything else that renders MCP prompts as slash commands) when you're already on a Tuple call and want the agent to silently follow along. The returned prompt: * Subscribes the agent to the unified events + transcript stream so it wakes on signal, not on a timer. * Maps participant IDs to names with `tuple state`. * Teaches the agent when to speak (direct address, terminal input, call end, mid-call checkpoint) and when to stay quiet. * Embeds the last `tail` lifecycle events and transcript lines so a mid-call join doesn't replay the whole history. If you'd rather launch a new agent process onto the call from your shell — instead of invoking it from an already-running agent — use [`tuple connect`](/cli/connect), which builds the same context and execs the harness for you. ```text theme={null} /tuple:connect /tuple:connect tail=200 /tuple:connect call_id=01HFE… tail=50 ``` ## Resources The server also exposes a set of [subscribable MCP resources](https://modelcontextprotocol.io/specification/server/resources). Agents can read them once for a snapshot, or subscribe for live updates. | URI | Content | | ----------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `tuple://state` | Full app state (current user, current call, contacts, rooms, websocket). Subscribable. This is the full-payload surface — `get_state` returns the bounded summary instead. | | `tuple://call/{callId}/transcription` | NDJSON transcription segments for a call. Subscribable. | | `tuple://call/{callId}/transcription.json` | Structured JSON snapshot of the same segments. | | `tuple://call/{callId}/events` | NDJSON transcription lifecycle events. Subscribable. | | `tuple://call/{callId}/events.json` | Structured JSON snapshot of the events. | | `tuple://call/{callId}/stream` | Merged NDJSON of events and transcript text. Subscribable. | | `tuple://call/{callId}/stream.json` | Structured JSON snapshot of the merged stream. | | `tuple://call/{callId}/notifications` | NDJSON notification events. Subscribable. | | `tuple://call/{callId}/notifications.json` | Structured JSON snapshot of notifications. | | `tuple://transcription/calls` | Index of past recorded calls (formatted text). | | `tuple://transcription/calls.json` | Structured JSON snapshot of the past-call index. | | `tuple://transcription/call/{callId}/transcript` | Full transcript of a past recorded call (formatted text). | | `tuple://transcription/call/{callId}/transcript.json` | Structured JSON snapshot of the past-call transcript. | Use `current` as `{callId}` to target the active call. Subscriptions deliver `ResourceUpdated` notifications when data changes; clients then re-read the resource for the new snapshot. The `tuple://transcription/...` resources are read-on-demand snapshots (not subscribable); the server also surfaces the most recent past calls as concrete, human-titled `transcription-…` resources so clients can attach one as context.