Which Claude Slack Setup Meets Your Delivery Contract?

Use Claude in Slack for coding tasks that should start a cloud session, standard Slack MCP for retrieval inside an initiated task, or a Channels bridge for a running session. Audit starts and replies against the original Slack thread.

Use a thread as the unit of acceptance

Start with two test threads containing identical requests and distinct synthetic markers. In a workspace eligible for the documented individual-account Claude in Slack flow, post a coding mention in each thread and record which cloud session was created under which account. For an organization using Claude Tag, write a separate expectation around the shared identity and its configured access. With standard Slack MCP, start the coding task independently, have it pull the permitted thread, and compare the retrieved marker. Count the lookup as a read and record any task start separately. If two threads contain the same request, a correct-looking response can still belong to the wrong conversation. Pair each result with a stable thread alias and require the marker from that thread before accepting the integration.

Claude Code in Slack - Claude Code Docs

Test the gap between receipt and visible reply

Claude Code Channels delivers events to an existing session and documents that delivery requires an open session. A bridge can add its own recovery behavior, which deserves a separate test. In an approved test workspace, send three synthetic events: one while the session is healthy, one during a planned test pause, and one replay of a previously handled event. Record times from a common clock and count task starts and visible replies independently. A duplicate reply with a single start points to a different stage than a duplicate start. For an illustrative 30-second deadline, an event replayed after 35 seconds is late even if the reply eventually lands. Give each transport the same event IDs and publish mismatch rows with the transport label so operators can compare the outcomes.

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

Test who starts the task

The official Claude app supports routing coding requests to Claude Code from Slack. Start the evaluation in a test channel with a repository approved for cloud use. Install or request the Anthropic app through the Slack Marketplace, then post a narrowly scoped coding request in a thread and tag Claude. Ask for a diagnosis and the relevant file path so you can check the result against the repository without making a change.

Before testing, check your workspace’s Claude Tag status and the identity your administrator expects to use. A useful pass is a coding task whose repository and account match that expectation. If you get a setup notice, a retirement notice, or a generic chat answer, record the response and resolve routing or eligibility before evaluating the code result.

Claude & Slack Integration | Slack Marketplace

Use Slack MCP for deliberate retrieval

Choose a standard Slack MCP connection when the operator starts Claude Code and decides when it should consult Slack. Anthropic’s Channels documentation describes standard MCP as a tool the session queries during a task. Before configuring a Slack MCP server, verify which authentication method and Slack scopes its current instructions require, and whether it offers the specific thread lookup you need. Grant access to the test data required for that lookup.

For an illustrative test, put the text sample-id 84B in a test thread and keep a fixture named parser-84B.json in your checkout. Ask Claude Code from the terminal to retrieve that thread and identify the matching fixture. A pass returns the exact identifier and correct path. Then post a second message while the session is idle. If that message is supposed to start work, your requirement calls for event delivery.

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

Qualify a bridge with approval and restart tests

Evaluate a bridge when Slack must start work on a host you operate or when the workflow needs a custom gate. The claude-code-slack-channel project documents a Slack app using Socket Mode and bot event subscriptions. Confirm that your Claude Code build supports Channels and, on Team or Enterprise, that an owner has enabled it. Configure the minimum event subscriptions for an isolated test and keep write or publish credentials out of the initial run.

If approval before publishing is a requirement, make it part of acceptance. Send a test mention, confirm the task reaches the intended session, and check that publication waits for approval. Stop the process, send a second numbered mention, and restart it. Record whether that message is dropped, delivered once, or repeated. Use the observed result to decide whether you need durable intake or duplicate handling.

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

Separate account failures from repository failures

For Anthropic’s earlier Claude Code in Slack integration, an individual session uses the requester’s connected repositories and plan limits. Claude Tag uses an organization’s shared identity with administrator-configured access. This matters when one person can open a coding task and another receives a notice or lands in the wrong repository. Record the workspace, requester, expected identity, and repository before changing permissions.

Use the same harmless request from two intended operators and check the resulting context for each. If neither request starts a coding task, investigate the workspace setup and the notice first. If the tasks start but only one can find an agreed file, check repository access for the identity in use. Keep a separate acceptance test for a local-only checkout or tool if either is required for the job.

Claude Code in Slack - Claude Code Docs

Sources