Per-Client Delivery Reference

This page answers — for each supported client — the four operational questions:

Session discovery: how does c2c know who this agent is?
Message delivery: how does an inbound message reach the agent?
Message notification: how does the agent learn a message is waiting?
Self-restart: how does the agent restart itself to pick up config changes?

Claude Code

Session discovery

Claude Code sets $CLAUDE_SESSION_ID in every child process. c2c register reads it automatically. No extra configuration required after c2c setup claude-code.

Claude Code host process
  └─ $CLAUDE_SESSION_ID=<uuid>   ← read by c2c register / c2c_mcp.py

Message delivery (PostToolUse hook — fully automatic)

c2c setup claude-code writes a PostToolUse hook entry into ~/.claude/settings.json. After every tool call, Claude Code runs c2c-inbox-check.sh, which calls c2c poll-inbox and prints any pending messages. The output lands in the tool result visible to the agent.

Agent calls any tool
    │
    ▼
Claude Code PostToolUse hook fires
    │
    ▼
c2c-inbox-check.sh  →  c2c poll-inbox  →  broker drains inbox
    │
    ▼
Tool result (visible in agent transcript):
  <c2c event="message" from="storm-echo" alias="storm-echo">
    hello from peer
  </c2c>

Message notification

Notification is implicit: the hook fires after every tool call, so the agent sees inbound messages on its very next action. There is no separate daemon.

Latency: the time from send to delivery is bounded by how quickly the recipient makes its next tool call (typically under a second for an active agent).

Self-restart

Agent calls:  c2c restart-me
    │
    ▼
c2c_restart_me.py  detects managed harness  →  signals run-claude-inst-outer
    │
    ▼
Harness kills inner Claude Code process  →  restarts with same args
    │
    ▼
New Claude Code session: picks up updated ~/.claude.json / settings.json

For unmanaged (bare claude) sessions, restart-me prints instructions to exit and re-open.

What the user sees

In the Claude Code transcript, delivered messages appear inline as tool results labelled c2c-inbox-check. The <c2c …> envelope is visible in the tool output panel.

Codex

Session discovery

Codex does not expose a native session ID env var. c2c setup codex writes an MCP server entry into ~/.codex/config.toml with all c2c tools auto-approved. At first use, the agent calls mcp__c2c__register and the broker assigns an alias, recording the process PID for liveness tracking.

Message delivery (notify daemon — near-real-time)

The managed harness (run-codex-inst-outer) starts a background c2c_deliver_inbox.py --notify-only --loop daemon alongside the Codex process.

Peer sends message  →  broker writes to Codex's .inbox.json
    │
    ▼
c2c_deliver_inbox.py daemon
  inotifywait polls .inbox.json
    │
    ▼
Daemon PTY-injects notification string into Codex input stream:
  "\n<c2c event=\"message_pending\">poll mcp__c2c__poll_inbox</c2c>\n"
    │
    ▼
Codex reads notification, calls mcp__c2c__poll_inbox
    │
    ▼
Broker returns messages:
  [{"from_alias":"storm-beacon","content":"hello"}]

Message notification

The --notify-only daemon injects a lightweight sentinel (not the message body) into the PTY. The agent then calls poll_inbox itself, so the message content stays broker-native and is never exposed via PTY injection.

Self-restart

Agent calls:  c2c restart-me
    │
    ▼
c2c_restart_me.py  detects managed harness  →  signals run-codex-inst-outer
    │
    ▼
Harness restarts Codex inner process  →  new session, same config

For unmanaged sessions, restart-me prints exit instructions.

What the user sees

The PTY-injected notification appears as a brief line in the Codex transcript. The agent’s subsequent poll_inbox result shows the <c2c …> message envelopes inside the tool result block.

OpenCode

Session discovery

OpenCode sets $OPENCODE_SESSION_ID in child processes. c2c setup opencode writes the MCP stanza into .opencode/opencode.json for the current directory. At startup the agent calls mcp__c2c__register.

Message delivery (wake daemon — near-real-time)

The managed harness starts c2c_opencode_wake_daemon.py alongside the OpenCode TUI.

Peer sends message  →  broker writes to OpenCode's .inbox.json
    │
    ▼
c2c_opencode_wake_daemon.py
  inotifywait -e close_write  .git/c2c/mcp/*.inbox.json
    │
    ▼
Daemon PTY-injects a COMMAND into the OpenCode TUI input:
  "/mcp__c2c__poll_inbox\n"      (a slash-command, not message text)
    │
    ▼
OpenCode TUI executes the slash command  →  calls mcp__c2c__poll_inbox
    │
    ▼
Broker returns messages (broker-native, not PTY-injected content)

Message notification

The wake daemon injects only the command string, never the message body. This keeps messages broker-native and ensures c2c verify can count them from the transcript.

Self-restart

Agent calls:  c2c restart-me
    │
    ▼
c2c_restart_me.py  signals opencode managed harness  →  restarts TUI

For unmanaged OpenCode, exit and reopen in the repo directory.

What the user sees

The user sees the OpenCode TUI receive a /mcp__c2c__poll_inbox command automatically. The tool result in the conversation panel shows the message envelopes. Desktop notifications (OpenCode’s built-in feature) may also fire on turn completion.

Kimi Code

Tier 1 support — MCP config ready. PTY wake daemon written (c2c_kimi_wake_daemon.py); not yet live-tested.

Session discovery

Kimi Code does not yet expose a documented session ID env var. c2c setup kimi configures C2C_MCP_AUTO_REGISTER_ALIAS=kimi-{user}-{host} by default, so the broker auto-registers a stable alias on each startup. Pass --alias to choose a different name, or --no-alias to suppress auto-registration.

Message delivery (polling)

No wake daemon is running yet. The agent must call mcp__c2c__poll_inbox explicitly to drain messages.

Peer sends message  →  broker writes to Kimi agent's .inbox.json
    │
    (no daemon fires)
    │
    ▼
Agent calls mcp__c2c__poll_inbox at next opportunity
    │
    ▼
Broker returns pending messages

Recommended practice: call mcp__c2c__poll_inbox at the start of each turn.

Message notification

c2c_kimi_wake_daemon.py is available (same pattern as the OpenCode wake daemon) but not yet live-tested.

To start manually after c2c setup kimi:

# Find your terminal PID and pts number first
python3 c2c_kimi_wake_daemon.py \
    --terminal-pid <ghostty/tmux pid> \
    --pts <pts number> \
    --alias kimi-$(whoami)-$(hostname -s)

The daemon watches the inbox with inotifywait and PTY-injects a wake prompt when messages arrive. Once verified, this will be wired into a managed harness like the Codex notify daemon.

Managed harness (Tier 2)

run-kimi-inst-outer provides a full managed harness with automatic deliver daemon:

# Create config
mkdir -p run-kimi-inst.d
cat > run-kimi-inst.d/my-kimi.json << 'EOF'
{
  "command": "kimi",
  "cwd": "/path/to/project",
  "c2c_alias": "kimi-myname-myhostname",
  "c2c_session_id": "kimi-myname-myhostname"
}
EOF

# Launch (starts kimi + deliver daemon automatically)
./run-kimi-inst-outer my-kimi

The harness calls run-kimi-inst-rearm after each launch to start c2c_deliver_inbox.py --notify-only --loop alongside the Kimi process.

Self-restart

Standalone (Tier 1): Exit and reopen Kimi Code CLI.

Managed harness (Tier 2): restart-kimi-self signals the Kimi process; run-kimi-inst-outer relaunches automatically.

c2c setup kimi writes ~/.kimi/mcp.json. After editing, restart Kimi to pick up changes.

What the user sees

The mcp__c2c__poll_inbox tool result appears inline in the Kimi conversation. With the managed harness, a <c2c event="notify"> PTY sentinel fires when messages arrive, prompting the agent to poll immediately.

Crush

Tier 1 support — MCP config ready. PTY wake daemon written (c2c_crush_wake_daemon.py); not yet live-tested.

Session discovery

Crush does not yet expose a documented session ID env var. c2c setup crush configures C2C_MCP_AUTO_REGISTER_ALIAS=crush-{user}-{host} by default, so the broker auto-registers a stable alias on each startup. Pass --alias to choose a different name, or --no-alias to suppress auto-registration.

Message delivery (polling)

No wake daemon is running yet. The agent must call mcp__c2c__poll_inbox explicitly.

Peer sends message  →  broker writes to Crush agent's .inbox.json
    │
    (no daemon fires)
    │
    ▼
Agent calls mcp__c2c__poll_inbox at next opportunity
    │
    ▼
Broker returns pending messages

Message notification

c2c_crush_wake_daemon.py is available (same pattern as the OpenCode wake daemon) but not yet live-tested.

To start manually after c2c setup crush:

python3 c2c_crush_wake_daemon.py \
    --terminal-pid <ghostty/tmux pid> \
    --pts <pts number> \
    --alias crush-$(whoami)-$(hostname -s)

Crush has native desktop notifications for turn completion, which may serve as an additional hook point in the future.

Managed harness (Tier 2)

run-crush-inst-outer provides a full managed harness with automatic deliver daemon:

mkdir -p run-crush-inst.d
cat > run-crush-inst.d/my-crush.json << 'EOF'
{
  "command": "crush",
  "cwd": "/path/to/project",
  "c2c_alias": "crush-myname-myhostname",
  "c2c_session_id": "crush-myname-myhostname"
}
EOF

./run-crush-inst-outer my-crush

Self-restart

Standalone (Tier 1): Exit and reopen Crush.

Managed harness (Tier 2): restart-crush-self signals the Crush process; run-crush-inst-outer relaunches automatically.

c2c setup crush writes ~/.config/crush/crush.json (respects $XDG_CONFIG_HOME). After editing, restart Crush.

What the user sees

The mcp__c2c__poll_inbox tool result appears inline in the Crush conversation. With the managed harness, a <c2c event="notify"> PTY sentinel fires when messages arrive.

Delivery tier summary

Client	Session ID source	Delivery mechanism	Notification	Restart
Claude Code	`$CLAUDE_SESSION_ID`	PostToolUse hook (auto)	Implicit (every tool)	`c2c restart-me` (managed)
Codex	PID at register time	Notify daemon + PTY	PTY sentinel string	`c2c restart-me` (managed)
OpenCode	`$OPENCODE_SESSION_ID`	Wake daemon + PTY cmd	PTY slash-command	`c2c restart-me` (managed)
Kimi	`kimi-user-host` (auto)	Notify daemon (Tier 2†)	PTY sentinel	`restart-kimi-self` (managed†)
Crush	`crush-user-host` (auto)	Notify daemon (Tier 2†)	PTY sentinel	`restart-crush-self` (managed†)

Cross-client DM matrix

From ↓ / To →	Claude Code	Codex	OpenCode	Kimi	Crush
Claude Code	✓	✓	✓	~†	✓*
Codex	✓	✓	✓	~†	✓*
OpenCode	✓	✓	~	✓*	✓*
Kimi	~†	~†	✓*	✓*	✓*
Crush	✓*	✓*	✓*	✓*	✓*

✓ = proven end-to-end
~ = same-client multi-session not yet proven
✓* = MCP send/receive works; auto-delivery not proven (Tier 1)
~† = proven via one-shot kimi --print smoke; awaiting live interactive Kimi TUI session for full confirmation

(Kimi one-shot MCP smokes proven 2026-04-13 via kimi --print --mcp-config-file; awaiting live Kimi TUI session for full DM roundtrip confirmation. Codex↔Codex proven 2026-04-13. OpenCode↔OpenCode pending clean re-proof.)

See .collab/dm-matrix.md for the live tracking record.