DEV Community

Cover image for Let Your AI Agent Use Slack: One CLI, No Slack App Needed
Shaharia Azam
Shaharia Azam

Posted on

Let Your AI Agent Use Slack: One CLI, No Slack App Needed

#slack #cli #agentskill #ai

Your AI agent can read your repo, run your tests, and open a pull request. Then it hits a wall. The context it needs is in Slack: the incident thread, the decision nobody wrote down, the "we changed the deploy order last week" message.

Getting an agent into Slack normally means building a Slack app, picking OAuth scopes, and waiting for an admin to approve it. That is a lot of work before the first useful message.

SlackCLI is a different path. It is a single open source binary that talks to Slack from your terminal, and every read command speaks JSON. If your agent can run a shell command, it can use Slack.

SlackCLI is an unofficial project. It is not affiliated with or supported by Slack Technologies.

Up and running in about a minute 

brew tap shaharia-lab/tap
brew install slackcli

slackcli auth login-auto
Enter fullscreen mode Exit fullscreen mode

login-auto opens a browser, you sign in to Slack the way you always do, and SlackCLI captures the session tokens for every workspace on that account. Nothing leaves your machine. Credentials land in ~/.config/slackcli/workspaces.json with file mode 0600.

Prefer a real bot token for a server or a CI job? That works too:

slackcli auth login --token=xoxb-your-token --workspace-name="My Team"
Enter fullscreen mode Exit fullscreen mode

Now try something:

slackcli conversations unread
Enter fullscreen mode Exit fullscreen mode

See it in action

Demo: https://github.com/shaharia-lab/slackcli#-see-it-in-action

Sign in, browse conversations, search the workspace, read a thread from a permalink, reply, react, read a Canvas as Markdown, and pipe --json into jq. All from the terminal.

Why a CLI is a good tool for an AI agent

Most agent frameworks are happiest when a tool is a plain command with plain output. SlackCLI is built exactly that way.

  • JSON everywhere. Every read command takes --json, so the agent gets structured data instead of screen scraping.
  • Clean streams. JSON goes to stdout. Spinners, warnings, and update notices go to stderr. A pipe carries only data.
  • Simple exit codes. 0 on success, 1 on failure. An empty search result is still a success, so check the data, not the exit code.
  • Links work as input. Anywhere the CLI wants a channel ID or a timestamp, you can paste a Slack permalink instead. A human drops a link in the prompt and the agent runs with it.
  • One static binary. No Python environment, no Node runtime, no server to keep alive.
  • Many workspaces at once. Add --workspace=automation-bot to any command to pick an identity on purpose.

Seven things an agent can do today

1. Catch up on what you missed 

slackcli conversations unread --json | jq '[.unread_channels[] | {name, unread_count}]'
Enter fullscreen mode Exit fullscreen mode

Feed that to a model and you have a morning digest.

2. Read a thread from a link 

slackcli conversations read --permalink="$LINK" --json | jq -r '.messages[].text'
Enter fullscreen mode Exit fullscreen mode

The JSON also carries a resolved users array, so user IDs are not opaque.

3. Search the workspace history 

slackcli search messages "deploy failed" --in=engineering --limit=50 --json
Enter fullscreen mode Exit fullscreen mode

All of Slack's own search operators work: in:, from:, before:, after:, has:, is:.

4. Find a channel or a person 

slackcli search channels incident --json
slackcli search people "ada@example.com" --json
Enter fullscreen mode Exit fullscreen mode

5. Reply where the conversation is happening 

slackcli messages send --permalink="$LINK" --message="Root cause found, fix is in #4821"
Enter fullscreen mode Exit fullscreen mode

Passing a permalink replies in that thread, so no ID juggling is needed.

6. Post a report humans actually want to read 

slackcli messages send \
  --recipient-id=C1234567890 \
  --message="Nightly build report" \
  --blocks='[{"type":"markdown","text":"# Nightly build\n\n- [x] Build\n- [x] Tests\n- [ ] Deploy"}]'
Enter fullscreen mode Exit fullscreen mode

Native Block Kit markdown and table blocks mean headings, task lists, code fences, and real tables instead of a wall of text.

7. Read a Canvas as Markdown 

slackcli canvas read F1234567890 --json | jq -r '.markdown'
Enter fullscreen mode Exit fullscreen mode

Team runbooks and specs often live in a Canvas. Now they are just Markdown your agent can read.

Wiring it into an agent

You do not need an SDK or an MCP server. Give the agent shell access and one short instruction block:

You can use the `slackcli` command to work with Slack.

Read commands (always add --json):
  slackcli conversations unread --json
  slackcli conversations read <channel-id|--permalink=URL> --json
  slackcli search messages "<query>" --in=<channel> --json
  slackcli canvas read <file-id> --json

Write commands (ask me first):
  slackcli messages send --permalink=<url> --message="<text>"
  slackcli messages react --permalink=<url> --emoji=<name>

Run `slackcli <group> --help` if you need the exact options.
Enter fullscreen mode Exit fullscreen mode

That is the whole integration. A useful first prompt:

Read the last 50 messages in #incidents, find any unresolved issue from today, and draft a summary for me. Do not post it yet.

Because slackcli <group> --help prints the authoritative options for the installed version, an agent can discover the rest on its own.

Keep it safe

Slack access is real access, so a few rules are worth setting up front.

  1. Separate the identity. Give automation its own bot token workspace profile instead of copying your personal browser session onto a server.
  2. Keep a human on writes. Reads are cheap and reversible. Posting is not. A simple "ask before sending" rule in the agent prompt goes a long way.
  3. Treat the config as a secret. ~/.config/slackcli/workspaces.json and the browser profile hold live credentials. Do not commit them, sync them, or copy them around. slackcli auth logout clears both.
  4. Refresh tokens without a window. Browser tokens expire with the session. slackcli auth login-auto --headless renews them in an unattended job once the profile has signed in once.
  5. Watch the rate limits. Commands that resolve many users or channels make one API call per entity, which adds up on a large workspace.

Try it 

brew tap shaharia-lab/tap
brew install slackcli
slackcli auth login-auto
slackcli conversations unread
Enter fullscreen mode Exit fullscreen mode

SlackCLI is MIT licensed, built with Bun, and ships prebuilt binaries for macOS, Linux, and Windows. It has over 400 tests and a full user guide.

⭐ Star it on GitHub if it saves you a trip to the Slack tab, and tell me in the comments what you would want your agent to do in Slack. Feature requests start as an issue.

Top comments (0)