Can Your Claude Code Channel Handle a Restart?

Use Claude Code Channels to push events into an open session and Remote Control to steer a session yourself. Test the chosen bridge with correlated probes across a restart before relying on it.

Use a bridge only where live context earns its keep

Channels are research-preview MCP integrations for pushing events into a running Claude Code session. Anthropic's excerpt documents claude.ai or Console authentication and explicit enablement for Team and Enterprise organizations. Check those prerequisites on the deployment you will use. In a test conversation, tell Claude a short marker and send a message through your chosen bridge asking it to return the marker and echo a probe ID. Compare that with Remote Control for a human steering an existing session, and with a fresh task for work whose ticket already contains all the context. The bridge is valuable when a push needs the running session's state. If a fresh task produces an equally usable answer from the ticket, you can avoid making delivery depend on that particular session being open.

Push events into a running session with channels - Claude Code Docs

Distinguish a reply from a completed round trip

For a Slack channel integration that documents both a policy layer and a decision journal, compare the sender's timestamp, the journal entry when applicable, and the visible reply using the same probe ID. The supplied Slack project describes thread isolation and a signed audit journal. Verify behavior in your deployment by opening two test threads, giving each a different harmless marker, and sending a probe in each thread before a restart, while the worker is stopped, and after it returns. Record the marker match in the correct thread and count repeated replies separately. A reply after recovery is useful evidence of delayed delivery only if its ID matches the earlier probe. A journal entry without a corresponding visible reply helps narrow the failure to a later stage. Keep observing for a fixed period past the reply deadline so delayed replies and missing replies can be compared consistently.

GitHub - jeremylongshore/claude-code-slack-channel: Slack-native governance substrate for Claude Code and peer agents: every tool call passes a tiered policy engine and lands in a hash-chained, Ed25519-signed audit journal you can verify offline. The governance kernel the companion agent-governance-plane (AGP) builds on. · GitHub

Decide whether the event needs this session

Choose a Channel when a message needs to reach the Claude Code session that already holds the relevant working context. A two-way bridge can also return a reply through the same channel. Before choosing a surface, write down what the first response must depend on, such as a decision made earlier in the session or a file in its current checkout.

For an illustrative test, start a session with two fictional release candidates, Cedar and Birch. Tell the session that Cedar is approved, then send a message from the intended channel asking which candidate was approved and requesting a reply without edits. Success is a reply naming Cedar in the same chat. A terminal event without a chat reply confirms only the inbound half of the route.

Push events into a running session with channels - Claude Code Docs

Match the surface to who starts the work

Use Remote Control when you need to pick up a local Claude Code session yourself from another device. Try it by opening a disposable session, giving it a distinctive piece of context, and confirming that you can steer that conversation from your phone. Choose a standard on-demand MCP tool when the task just needs a lookup that Claude can initiate during a turn.

For work that can begin with a complete brief, evaluate Claude Code on the web with a task that needs only a repository and written instructions. Consider Claude in Slack if the team wants to initiate work from an @Claude mention. Verify the Slack integration’s current availability and session behavior with a test mention. Check where the task appears, what workspace opens, and whether it depends on any context from your terminal.

Overview - Claude Code Docs

Qualify one messaging bridge

Start with the platform the operator will actually use. Telegram, Discord, and iMessage are documented event sources. Pick one and use the current channel instructions for that plugin’s prerequisites, credentials, and sender authorization. Confirm that the intended account is on the approved sender list before you leave the session unattended.

Test from the actual device and account that will send requests. Send a unique, harmless message and confirm that it lands in the intended session. For a two-way workflow, also check the reply in the originating chat. Where you can safely test with a second account you control, verify that an unapproved sender cannot trigger the same action. If a team shares Discord, run the test in the intended room so the result covers its actual access settings.

Overview - Claude Code Docs

Set a limit on missed messages

Channels receive events while the Claude Code session is open. Decide how long an event can go unacknowledged before someone intervenes. For a discretionary question, a missed reply can wait until you return. For a release decision with a deadline, keep the original request in a system that people can check even when the session is down.

Run two probes in a test workspace. Send one with the session open, then close it and send another with a different identifier. Reopen the session and send a third. The first and third probes establish normal delivery and recovery. Check the second separately to learn how your bridge handles interruptions. When a fresh post-restart probe fails, check the session and plugin before investigating the task prompt.

Push events into a running session with channels - Claude Code Docs

Sources