Slack MCP server: connect AI agents to Slack from the EU

Slack, owned by Salesforce and based in San Francisco, is the chat platform many companies use for team channels, direct messages and shared documents called canvases. Its official MCP server runs at https://mcp.slack.com/mcp over Streamable HTTP with JSON-RPC 2.0.

The server acts with a user token, so every search, read and post happens as the person who authorised it, within their channels. Slack applies the same rate limits as its Web API per tool, and MCP activity appears in Slack's audit logs.

A project agent can answer "what did we decide about the pricing page?" by searching Slack history and quoting the thread with a link. Each Friday it can read a project channel, write a summary canvas and post it to the leadership channel after review.

Support teams can have an agent watch for questions in a help channel, look up answers in documentation connectors and draft replies. Engineering teams can let an agent post a release note to a channel once a GitHub release is published.

Slack requires every MCP client to be backed by a registered Slack app; only Marketplace apps and internal apps may use MCP. For an inference.eu agent, your workspace creates an internal Slack app with the user scopes the agent needs, such as search:read.public or chat:write.

Add the Slack connector on inference.eu, enter that app's client ID and secret, then sign in with Slack through OAuth. Credentials and tokens are stored encrypted in the EU. Workspace admins approve the app through Slack's usual app approval process.

Messages are fetched from Slack's own MCP endpoint, so data stays with Slack, a US company, as it does for your other Slack integrations. Slack Enterprise customers who use data residency keep their workspace data in the region they chose.

On inference.eu, the model that reads messages runs in EU data centres, and your Slack credentials are stored encrypted in the EU. Slack history contains personal and confidential conversations, so request only the scopes the agent needs and avoid private and direct message scopes unless there is a clear reason.

Sending a message is visible and cannot be unsent quietly, so posting should usually need a person's approval, at least at first. Reading channels and searching are lower risk but still expose private discussion to the agent.

On inference.eu you can require approval for every send, schedule, channel creation and canvas update, and let searches run on their own. Slack admins can see and revoke the app, apply IP allowlists to it, and review MCP actions in the audit logs.