Slack's hosted MCP server. The endpoint, the app-identity rules that decide who may connect, the OAuth flow and the user-token scopes behind each tool, read from the Slack developer documentation on 2 September 2026.
Interactive examples · No account is connected on this page
Updated
Slack MCP server is Slack's own Model Context Protocol endpoint at https://mcp.slack.com/mcp. It speaks JSON-RPC 2.0 over Streamable HTTP, and its tools search messages, files, users and channels, read channels and threads, send messages, manage canvases and lists, and upload files. Slack supports neither SSE nor Dynamic Client Registration, so every client has to be backed by a registered Slack app.
Pick an example to see which Slack MCP tools a task would call and what the result looks like. Examples are illustrative; nothing on this page connects to Slack.
Slack's MCP server lives at https://mcp.slack.com/mcp and speaks JSON-RPC 2.0 over Streamable HTTP only. Every client must be backed by a registered Slack app with a fixed app ID, so the simplest route is a partner client: Claude.ai, Claude Code, Perplexity or Cursor. Your own client needs confidential OAuth and only the user-token scopes its tools require.
This is the step that catches people out. Slack's documentation says MCP clients must be backed by a registered Slack app with a fixed app ID, and that the client has to hardcode that ID. Only directory-published apps or internal apps may use MCP; unlisted apps are prohibited. That is what lets workspace admins approve and manage the client through the normal Slack app approval process.
https://mcp.slack.com/mcpSlack lists four clients where the server is already available with no coding: Claude.ai, Claude Code, Perplexity and Cursor. Each of those ships its own registered Slack app, so you connect from inside the client and approve it in your workspace. This is the shortest path, and the only one that does not require you to run through Slack app registration yourself.
https://claude.ai
https://code.claude.com
https://perplexity.ai
https://cursor.comAll requests go to one URL over Streamable HTTP. The documentation is explicit that SSE-based connections are not supported at this time, so a client that can only speak SSE cannot connect. Dynamic Client Registration is not supported either, which is why the fixed app ID above matters.
https://mcp.slack.com/mcpSlack supports confidential OAuth for MCP clients, using your app's client_id and client_secret. If your client implements OAuth 2.0 Authorization Server Metadata (RFC 8414), it can discover everything from the two well-known documents Slack publishes. PKCE support is available for desktop clients.
https://mcp.slack.com/.well-known/oauth-protected-resource
https://mcp.slack.com/.well-known/oauth-authorization-server
Authorize: https://slack.com/oauth/v2_user/authorize
Token: https://slack.com/api/oauth.v2.user.accessScopes are granted per tool, not per server, so a read-only integration never has to ask for write access. Search needs search:read.public plus the private, mpim and im variants for those conversation types. Reading channels and threads needs the four history scopes. Sending needs chat:write, reactions need reactions:write, canvases need canvases:read and canvases:write, lists need lists:read and lists:write, and uploads need files:write.
search:read.public search:read.private search:read.mpim search:read.im
channels:history groups:history mpim:history im:history
chat:write reactions:write files:read files:write
canvases:read canvases:write lists:read lists:write
users:read users:read.email channels:read groups:readTwo settings can silently break a working config. If your app has allowed IP address ranges configured, requests to the MCP server are subject to the same restriction and calls from other addresses are rejected, so the client's egress addresses have to be on the list. Separately, MCP activity is recorded in Slack's audit logs, which is where to look when you need to see what an agent actually did.
Audit Logs API — actions reference, mcp-server sectionOfficial documentation: Read the Slack MCP docs.
There are two ways to reach the Slack MCP server, and both use the same endpoint and the same admin approval. Through a partner client you build nothing, because the client vendor ships a registered Slack app and holds the OAuth credentials. Through your own Slack app you manage client_id and client_secret yourself and choose the scopes, so read-only access is achievable.
| What differs | Through a partner client | Through your own Slack app |
|---|---|---|
| What you build | Nothing; the client ships a registered Slack app | A Slack app with a fixed app ID, published to the Marketplace or internal to your workspace |
| Clients Slack names | Claude.ai, Claude Code, Perplexity, Cursor | Any MCP client you control that speaks Streamable HTTP |
| Endpoint | https://mcp.slack.com/mcp | https://mcp.slack.com/mcp |
| OAuth credentials | Held by the client vendor | Your app's client_id and client_secret; PKCE available for desktop |
| Who approves it | Workspace admins, through the standard Slack app approval process | Workspace admins, through the same process |
| Scope control | Whatever the client requests | You choose the user-token scopes, so read-only is achievable |
Endpoint, transport, app-identity rules, scope table and client list transcribed from the Slack MCP server overview at docs.slack.dev, read 2026-09-02.
Inside a workspace the Slack MCP server can search messages, files, users and channels, read full channel and thread histories, send messages, create and read canvases as markdown, work with Slack lists, and upload files in a deliberate two-step flow. Rate limits are the same per-method limits as the Slack Web API.
Messages and files, filtered by date, user and content type. Users, by name with partial matching, email or user ID. Public and private channels, by name and description. Custom emoji. There is also a channel listing that returns the conversations you are a member of, filterable by type and name prefix, with pagination and archived channels included by default.
Complete channel histories and complete thread conversations, sending to any conversation type, and drafting and previewing a message inside the AI client before it goes out. It can also create a channel, group DM or IM on behalf of the authenticated user, and add emoji reactions.
Create and update Slack canvases, and read one back exported as a markdown file. That export is the practical route for getting a formatted Slack document into a model's context without screenshots.
Uploads are deliberately not a single call. Call slack_get_file_upload_url to get a signed URL and a file ID, POST the bytes to that URL, then call slack_complete_file_upload with the file ID to finalise it and optionally share it to a channel. The file is not visible until the second step completes.
Lists hold structured tabular data with typed columns — text, date, select, user and more. The server can create lists with custom schemas, read their contents, update list metadata and columns, and add or update individual records. Columns are addressed by display name or internal key.
Limits are enforced per tool or action type, and are the same whether the call arrives through MCP or a direct Web API method. Reading a channel or thread sits at Tier 3, searching users, channels or emoji at Tier 2, reading a user profile or listing channel members at Tier 4, while search and send-message have their own special limits.
Slack MCP server is Slack’s official Model Context Protocol endpoint at https://mcp.slack.com/mcp. It speaks JSON-RPC 2.0 over Streamable HTTP and lets an AI client search and read messages, files, users and channels, send messages, and manage canvases and lists.
Slack does not support SSE or Dynamic Client Registration, so every client is backed by a registered Slack app — either a partner client Slack has approved or your own app with confidential OAuth.
Slack MCP brings conversation history into an agent. The guides below cover the other systems the same agent usually has to touch once it has read a thread: a knowledge base, code hosting, broad app connectors and a browser. Each guide lists install steps, authentication and the tools each server exposes.
Notion MCP ServerNotion's hosted server for pages and databases.
GitHub MCP ServerRepositories, issues, pull requests and CI logs.Zapier MCP ServerZapier actions exposed to an agent as tools.
Claude MCPAdding MCP servers to Claude clients.
Cursor MCPHow Cursor loads MCP servers from mcp.json.n8n MCP Servern8n workflows exposed as MCP tools.Playwright MCP ServerBrowser automation for sites with no API.Every client needs a registered Slack app; partner clients have one already.
For your own app, configure confidential OAuth and request only the user-token scopes your tools need.
Search or read one channel before allowing the agent to send messages.
Read a channel’s history and its threads, then summarise decisions and open questions.
Search messages and files filtered by date, user and content type.
Create or update a canvas from the summary, and export one back as markdown for the model.
Add or update records in a Slack list with typed columns.
These answers cover the Slack MCP constraints that surprise people: whether any client can connect, which clients work without code, whether SSE is supported, how to give an agent read-only access, why calls fail after OAuth succeeded, and where to see what an agent did in Slack.
No, and this is the constraint that surprises people. Slack's documentation says MCP clients must be backed by a registered Slack app with a fixed, hardcoded app ID, and that only directory-published apps or internal apps may use MCP — unlisted apps are prohibited. Dynamic Client Registration is not supported either. So a generic client that expects to register itself on the fly cannot connect. Either use one of the four clients Slack names, or register your own Slack app.
Slack's documentation lists four partner-built clients: Claude.ai, Claude Code, Perplexity and Cursor. Each ships its own registered Slack app, so you connect from inside the client and your workspace admin approves it through the standard Slack app approval process.
No. Slack supports JSON-RPC 2.0 over Streamable HTTP only, with all requests sent to https://mcp.slack.com/mcp, and the documentation states plainly that SSE-based connections are not supported at this time. If your client offers a transport choice, pick Streamable HTTP.
Effectively, yes, if you control the app. Scopes are granted per tool on the user token, so an app that requests only the search and history scopes — search:read.public, search:read.private, channels:history, groups:history, mpim:history, im:history, plus files:read and users:read — has no way to post. Leave out chat:write, reactions:write, files:write, canvases:write and lists:write and the writing tools have nothing to authenticate with.
Check two things the documentation calls out. If your app has allowed IP address ranges configured, requests to the Slack MCP server are subject to the same restriction, and calls from addresses outside the list are rejected — so the client's egress addresses must be on it. Second, Slack MCP is subject to the same Web API rate limits as direct method calls, enforced per tool, so a burst of searches can hit a Tier 2 limit while the rest of the server keeps working.
Slack records MCP activity in its audit logs; the Audit Logs API actions reference has a section for the MCP server. That is the trail to check when you need to reconstruct which messages were sent or which content was read, rather than relying on the AI client's own transcript.
Pulling a decision out of a thread is one call. Turning it into a ticket, a record update and a follow-up on a schedule is a workflow, and that is a different piece of the stack. It needs a trigger, a place to write the result, and a rule for which messages an agent may post on its own.