Summary

It reads the terminal, which is why it can supervise rather than just launch.

Most multi-agent setups fire off a subprocess and wait for output. This one emulates the terminal and parses state, tool output, the status bar and confirmation dialogs, so the orchestrator can answer a permission prompt, notice a stuck worker from its JSONL activity, or restart a slot before its context runs out. That is also the cost: a daemon, a Postgres instance and a blueprint file before the first agent runs.

What it is

A local orchestration daemon that spawns coding agents inside pseudo-terminals and coordinates them from your main client. It parses each agent's terminal in real time — idle, thinking, responding, confirming, error — so the orchestrator knows what a worker is doing rather than guessing, and it routes work through explicit intent and plan approval before anything executes. Postgres is the runtime store and event log.

What you get

PTY control: `mission_pty_spawn`, `mission_pty_send`, `mission_pty_screen`, `mission_pty_confirm` for tool dialogs, `mission_pty_interrupt`, `mission_pty_kill`, `mission_pty_status`, `mission_pty_history`, `mission_pty_logs`, and `mission_pty_screenshot`, which renders the terminal as a PNG
Task operations: `mission_submit`, `mission_ask` for a synchronous consultation, `mission_status`, `mission_cancel`, `mission_spawn`, `mission_kill`, `mission_restart`, `mission_agents`, `mission_slots`, `mission_inbox`
A knowledge base: `mission_kb_remember`, `mission_kb_search` over full text, `mission_kb_get`, `mission_kb_list`, `mission_kb_import`, `mission_kb_discover`, `mission_kb_gc`, `mission_kb_analyze`, `mission_kb_forget`
A task board: `mission_board_create`, `mission_board_list`, `mission_board_update`, `mission_board_get`, `mission_board_delete`, `mission_board_toggle`, `mission_board_note_add`, `mission_board_summary`
Cross-session monitoring: `mission_cc_sessions`, `mission_cc_tasks`, `mission_cc_overview`, `mission_cc_in_progress`, `mission_slot_history`
Conversation and memory: `mission_conversation_list`, `mission_conversation_get`, `mission_conversation_search`, `mission_memory_pending`, `mission_memory_pause`, `mission_memory_done`, `mission_token_stats`
Permissions per role and per slot: `mission_permission_get`, `mission_permission_set_role`, `mission_permission_set_slot`, `mission_permission_add_auto_allow`, `mission_permission_reload`, with allow, confirm and deny outcomes matched by glob
A question queue so a blocked agent posts a question for a human instead of stalling: `mission_question_create`, `mission_question_list`, `mission_question_get`
Auto-restart when a slot's context window drops below 10%, and stuck detection from JSONL activity

Requirements

Install the binaries with `cargo install missiond-mcp --bin mission-mcp`, `cargo install missiond-daemon --bin missiond` and `cargo install missiond-attach --bin missiond-attach`, or take the prebuilt binaries from npm as `@missiond/core` for macOS, Linux and Windows. Register `mission-mcp` as the command in your client config. Postgres is the runtime store and event log, so you need one available. Slot configuration — which roles exist and where they work — comes from the project's blueprint file.

Setup effort

One command — cargo install missiond-mcp --bin mission-mcp