Slack MCP Server: Official OAuth & Self-Hosted Setup

What Is a Slack MCP Server?

A Slack MCP server connects an AI client to Slack through the Model Context Protocol. The important distinction is between Slack’s official hosted server, connected through OAuth, and community or self-hosted implementations, which may use bot tokens and expose different tools.

Use the official Slack MCP documentation to check supported clients, tools and access requirements. Do not apply a community server’s npm commands or bot-token configuration to the official connector.

Why Connect Slack to AI via MCP?

Useful starting points are searching conversations, summarising a thread and preparing a reply. For example: “Find the discussion about the quarterly review and summarise the decisions with links to the source messages.”

Results depend on the connected user’s access and the tools available in that client. A connector does not grant unrestricted access to every public channel, private channel or direct message.

Setup Method 1: The Official Hosted Slack Connector

For Claude, follow Slack’s official connection guide. Client availability, account requirements and workspace policies can change; use that guide for the current setup flow.

1. Check that your client and workspace meet the documented requirements. If your organisation controls app installation or connectors, arrange the required approval first.

2. Choose the Slack connector through the client’s supported setup flow.

3. Sign in to the intended Slack workspace and review the OAuth permissions before authorising the connection.

4. Start with a read-only request against a conversation you can already access in Slack.

5. Review the returned sources. Test any write capability separately and confirm the destination and message before sending.

This uses a hosted OAuth connection. It does not require installing an npm package or copying a bot token into a local configuration file.

Setup Method 2: Community or Self-Hosted Implementations

A separate implementation can be useful when you need custom tooling or deployment control. It is not interchangeable with the official Slack connector.

Choose an implementation first, then follow its maintained documentation for installation, runtime version, authentication and transport. Verify the package owner, repository and supported tools before running a command. The npm examples previously shown here were not substantiated, so they have been removed; that is not a claim that a similarly named package cannot exist.

If the implementation requires a Slack app, use Slack’s app management and request only the permissions required by that implementation. Bot and user tokens have different capabilities. Do not assume that adding a search scope to a bot token provides the same search access as the official OAuth connector.

Keep tokens out of source control, follow workspace app-approval rules and test access with the intended account. Use the implementation’s documented configuration rather than a generic token or transport example.

Available Tools and Message Drafts

Tool availability depends on the server, client and granted permissions. Check the current official tool reference or your chosen implementation’s documentation before relying on search, channel and thread reading, message sending, scheduling or canvas operations.

A draft is not automatically a Slack Draft. The official documentation describes drafting in the AI client. Review the proposed text there; do not assume it appears in Slack’s Drafts view. Sending or scheduling is a separate action and should be reviewed before execution.

Real-World Use Cases

1. Daily Standup Summary

Prompt: “Summarise the #engineering discussion from the last 24 hours. Include blockers and links to the relevant messages. Do not post anything.”

2. Customer Feedback Aggregation

Prompt: “Find accessible conversations mentioning customer feedback or bug reports from the past week. Group the findings by product area and cite the original messages.”

3. Meeting Follow-Ups

Prompt: “Read this product-planning thread and draft a checklist of action items here for my review. Do not send it yet.”

These requests begin with retrieval and review. Automated notifications need an additional trigger or scheduling workflow; simply connecting MCP does not start background automation.

Security, Permissions and Troubleshooting

Missing connector or app approval: Check the official client requirements and your organisation’s connector and app-installation policy. Workspace approval and your own OAuth authorisation are separate checks.

Missing messages or empty search results: Confirm that the connected account can access the conversation in Slack. For a community server, also check its token type, required scopes, channel membership and supported search API. Official user-authorised access and bot membership are different models.

Message sending fails: Verify that the tool exists in the connected client and that its required permissions and destination access are present. Do not fix a failed write by granting broad permissions without checking the cause.

Rate limits: Limits depend on the Slack API method and app context. Follow Slack’s response and retry guidance, including Retry-After where returned. Automatic backoff is implementation-dependent; do not assume every MCP server retries for you.

Is the native connector enough?

Start with the task. If you need to find and summarise Slack conversations, first check whether the native connector in your chosen AI tool meets your access and workflow requirements.

For a process spanning Slack and other systems, map what it must read, decide and change. Define permitted actions and where a person must review or approve the result. Use an existing connector when it covers those requirements; consider a scoped integration for any remaining gaps.

I work directly with you to define and build the workflow, using AI to support research, analysis and planning. My AI integration services explain the approach.

How to Set Up a Slack MCP Server: Connect AI to Your Workspace