Getting an AI agent into Slack is the moment automation stops feeling like a science project and starts feeling like a teammate. Instead of opening a separate tool, someone types a question in a channel and gets an answer back in seconds. The good news: you do not need to write code. The slightly annoying news: Slack has two different ways to connect, and picking the wrong one is the single most common reason people get stuck. This guide walks through both, shows you exactly which to choose, and flags the gotchas we hit constantly when wiring these up.
First decide: does your agent only talk, or does it also listen?
Before touching any settings, answer one question, because it determines everything that follows.
- One-way (agent only sends messages): Your agent posts into Slack but never reads or responds to what people type. Think daily summaries, lead alerts, “a form was submitted” pings. This is the easy path.
- Two-way (agent listens and replies): Someone @mentions the agent or messages it, and it answers. This is a real conversational assistant and needs more setup.
The reason this matters: one-way only needs an Incoming Webhook, which is essentially a secret URL you POST a message to. Two-way needs a proper bot user with an OAuth token and event subscriptions. People try to build a two-way assistant on a webhook, hit a wall, and assume no-code “can’t do it.” It can — they just started on the wrong road.
| What you need | Use this | Setup effort |
|---|---|---|
| Post alerts / summaries into a channel | Incoming Webhook | ~5 minutes |
| Agent replies when @mentioned | Bot token + Events | ~20-30 minutes |
| Read messages, react, manage channels | Bot token + extra scopes | ~30 minutes |
The fast path: one-way alerts with an Incoming Webhook
If your agent just needs to broadcast, this takes minutes and you can stop reading after this section.
- Go to api.slack.com/apps and click Create New App → From scratch. Name it and pick your workspace.
- In the left sidebar, open Incoming Webhooks and toggle it on.
- Click Add New Webhook to Workspace, choose the channel to post in, and authorize.
- Slack hands you a URL like
https://hooks.slack.com/services/T000/B000/xxxx. Copy it. - In your no-code tool, add an HTTP Request step (or a native “Slack” action) that sends a POST to that URL with a JSON body like
{"text": "Your message here"}.
That is the whole thing. One honest caveat: a webhook is locked to one channel and can only push messages out. It cannot read replies, cannot DM people dynamically, and cannot react. If you ever need those, you have outgrown webhooks — go to the next section.
The real assistant: a two-way bot your agent can talk through
This is what most people actually want: type “@Assistant summarize this thread” and get a reply. Here is the setup that works.
Step 1 — Create the app and add bot scopes
From the same api.slack.com/apps page, create an app, then open OAuth & Permissions. Under Bot Token Scopes, add permissions based on what the agent will do. Add only what you need — Slack will make you re-authorize every time you change this, and over-permissioned apps are harder to get approved in stricter workspaces.
chat:write— send messages (you almost always need this)app_mentions:read— see when someone @mentions the botchannels:history— read messages in public channels (needed to summarize threads)channels:read— see channel infoim:historyandim:write— if you want the agent to work in direct messages
Step 2 — Install the app and grab the bot token
Scroll up on the same page and click Install to Workspace, then authorize. You will get a Bot User OAuth Token that starts with xoxb-. This token is the bridge between Slack and your no-code platform — treat it like a password. Paste it into your tool’s Slack credential field.
Step 3 — Connect your no-code platform
The exact clicks differ by tool, but the shape is identical: create a Slack credential using that xoxb- token, then build two halves — a trigger that fires when the bot is mentioned, and an action that posts the agent’s reply back to the same channel and thread.
- n8n: Use the Slack Trigger node set to “App Mention,” then a Slack node to send the reply. One critical gotcha: the trigger needs an “API Access Token” credential (the
xoxb-token), not the OAuth2 credential. OAuth2 works for the action node but not the trigger. This trips up almost everyone the first time. - Make: Use the Slack module’s “Watch Public Channel Messages” or the events trigger, then a “Create a Message” action.
- Zapier: “New Mention” trigger → your AI step → “Send Channel Message” action. Simplest to set up, but see the cost note below.
Step 4 — Tell Slack where to send events (the part people forget)
For mentions to reach your workflow, Slack needs a public URL to push events to. In your app settings, open Event Subscriptions, turn it on, and paste your workflow’s webhook/trigger URL (n8n and Make show this on the trigger node once you activate the workflow). Slack sends a one-time “challenge” to verify the URL — most no-code triggers answer it automatically, but the workflow must be active/published, not in test mode, or verification fails. Then under Subscribe to bot events, add app_mention. Save.
Step 5 — Insert the AI and test in a real channel
Between trigger and reply, drop your AI step (OpenAI, Anthropic, or a built-in AI node) and feed it the message text plus any context you pulled. Then invite the bot to a channel with /invite @YourBot and @mention it. A bot can only see and post in channels it has been invited to — “it’s not responding” is, nine times out of ten, a bot that was never invited.
The gotcha that will bite you in production: token rotation
Slack offers a security feature called token rotation that makes tokens expire every 12 hours. It sounds responsible, and for some setups it is — but most no-code Slack credentials that store a static xoxb- token will simply stop working once it rotates, and your agent goes silent overnight with no obvious error. For a production agent, leave token rotation off unless your platform explicitly supports refreshing rotated tokens. We have watched more than one “it worked yesterday!” mystery trace back to exactly this.
When Slack-native is NOT the right call
Being honest about fit saves you days. Skip the build-your-own route if:
- You only need a handful of simple alerts. A plain Incoming Webhook beats a full bot. Do not add OAuth, events, and an AI loop to send a “new signup” message.
- Your workspace is locked down by IT. Many companies require admin approval to install custom apps. If you can’t get it, a sanctioned tool like a Workflow Builder step or an approved connector may be your only path — fighting the admin policy is a losing game.
- You expect heavy volume on Zapier. Zapier is the friendliest to set up, but it bills per task. A busy channel where the agent replies hundreds of times a day gets expensive fast. At that volume, n8n (self-hosted, flat cost) is dramatically cheaper, just with a steeper first hour.
FAQ
Do I need to pay for Slack to add an AI agent?
No. Incoming webhooks and bot apps work on Slack’s free plan. Your cost comes from elsewhere: the AI model usage (per token/per call) and your automation platform if it charges by task, like Zapier or Make. n8n self-hosted avoids the per-task fee entirely.
Why does my bot post messages fine but never respond to mentions?
Almost always one of three things: the bot was never invited to the channel (/invite @YourBot), Event Subscriptions isn’t enabled with the app_mention event, or your workflow is in test mode so Slack’s URL verification challenge failed. Sending works on the token alone; listening requires the events plumbing on top. Check those three in order.
Is a no-code Slack agent secure enough for a company workspace?
It can be, with discipline. Store the xoxb- token in your platform’s credential vault (never paste it into a prompt or a public step), grant only the scopes the agent actually uses, and be deliberate about channels:history — that lets the agent read channel content, which may include sensitive data your AI provider then sees. For regulated data, confirm your model provider’s data policy before going live.
Next step
Start with the smallest version that proves value: wire up a one-way webhook that posts a single useful message — a morning summary or a “new lead” alert — into one channel today. Once your team sees the agent show up where they already work, upgrading it to a two-way assistant that answers questions is a natural, low-risk next move. Build the webhook first, earn the trust, then add the brain.