Managed Deep Agents is in public beta and available on LangSmith Cloud in the US region only.
Project structure
Slack setup uses a channel declaration, an editable manifest template, and a generated manifest:Add a Slack channel
Export a channel created withchannels.slack():
channels/slack.py
slack and mounts its Events API route at /channels/slack/events. You can use another file name when you need a different configured name.
Create and deploy the Slack app
After setting up the Slack channel, you need to create and deploy your Slack app.1
Deploy your Managed Deep Agent
First, deploy your Managed Deep Agent.Wait for the deployment to finish. The agent is deployed even though Slack is
not active yet.
2
Generate the app manifest template
Run a command to generate the Slack app manifest template.MDA finds the existing deployment and writes two files:
- A “template” manifest to
slack-app-manifest.json - The full manifest to
.mda/slack/app-manifest.json
mda channel add slack . to regenerate the full template.
Do not directly edit the generated file at .mda/slack/app-manifest.json3
Create and install the Slack app once
- Open https://api.slack.com/apps.
- Select Create New App.
- Select From an app manifest.
- Choose the target Slack workspace.
- Import
.mda/slack/app-manifest.jsonand create the app. - Open OAuth & Permissions and select Install to Workspace.
- Approve the requested permissions.
- Copy the Bot User OAuth Token from OAuth & Permissions.
- Copy the Signing Secret from Basic Information → App Credentials.
4
Add the Slack credentials
Add the two values to the project .env:Do not commit .env and do not copy these values into either manifest.
5
Redeploy to activate Slack
6
Updating Slack Bot Configurations
The previous steps guide you through deploying the first iteration of your Managed Deepagent as a Slack bot. If you want to make any updates to the agent, you just simply rerun
mda deploy . to update the deployed app.If you want to make any configuration changes to the Slack application itself, the recommended steps are:- Update the
slack-app-manifest.jsonfile locally - Run
mda channel add slack .— this will regenerate the manifest with the latest changes. These changes are written to the.mda/slack/app-manifest.jsonfile - On https://api.slack.com/apps → locate your app → App Manifest
- Replace the contents of the manifest with the new
.mda/slack/app-manifest.jsonfile → Click Save Changes - Navigate to the OAuth & Permissions tab. Click on Reinstall to Workspace
slack-app-manifest.json as the source of truth. When you change its scopes, bot events, branding, or other settings, rerun mda deploy . --configure-slack, apply the regenerated .mda/slack/app-manifest.json, and reinstall the app if Slack requests it. The generated files under .mda/ are build artifacts; do not commit them.
Configure Slack behavior
Pass options tochannels.slack(...) to control Managed Deep Agents runtime behavior. Configure OAuth scopes and delivered event types in the Slack app, not in the channel declaration.
channels/slack.py
Conversation mappings accept:
"thread": Reuse one Managed Deep Agents thread for a Slack thread."conversation": Reuse one Managed Deep Agents thread for the Slack conversation."message": Start a separate Managed Deep Agents thread for each message.
Understand event and thread behavior
The Slack app controls which events reach the deployment. The Slack channel normalizes supported events and applies the configured filters and conversation mapping.
Top-level channel messages that do not mention the bot are ignored. Bot messages, the app’s own messages, unsupported message subtypes, and events rejected by channel filters do not start runs.
When Slack delivers the same mention through both
app_mention and message.*, Managed Deep Agents drops the duplicate message event. Subscribe to app_mention for mentions and use message.channels or message.groups for non-mention follow-up replies.
Send responses to Slack
Withauto_reply enabled, Managed Deep Agents extracts the final assistant response and posts it to the originating Slack conversation after the run completes.
Channel-originated runs also expose runtime.channel in tools and middleware. Use it to inspect the normalized event, post an intermediate or final message, or update a previously posted message. It is absent for ordinary HTTP and scheduled runs.
The following tool posts the final response explicitly:
tools/send_channel_reply.py
{"final": True} only when the posted message is the final response. It suppresses the automatic reply so the user does not receive the final response twice. A post without that option is an intermediate message and does not suppress auto-reply.
runtime.channel.post(...) can post only to the originating Slack thread. Explicit destinations are not supported for channel-originated runs. To send a scheduled result to a specific Slack conversation, use deliver_to.
Understand Slack caller identity
A Slack event runs as an identity derived from the Slack workspace and user, such asslack:T123:U456. This identity is separate from caller identities used for HTTP requests. Slack account linking is not supported.
Deploy changes
Redeploy after changing the channel declaration, secrets, or identity configuration. Include--configure-slack when you change slack-app-manifest.json, the channel name, or the deployment so MDA can regenerate the final manifest with the current Events URL. Apply the generated manifest to the existing Slack app after the deployment completes.
Avoid making lasting configuration changes only in the Slack dashboard. A later manifest update can replace settings that are not present in the checked-in template.
Review security and current limits
- Managed Deep Agents verifies every Slack request against its raw body and rejects signatures outside Slack’s five-minute replay window.
- Slack Connect shared conversations are not supported.
runtime.channeldoes not exposeSLACK_BOT_TOKENor other provider credentials.- Event deduplication is currently process-local. A multi-replica deployment can invoke the agent more than once when Slack retries an event.
Troubleshoot Slack channels
--configure-slackreports that it needs exactly one channel: Keep exactly onechannels.slack(...)declaration in the project when using the manifest workflow.- MDA cannot read the template: Confirm
slack-app-manifest.jsonis a regular JSON file at the project root. Remove credentials and anysettings.event_subscriptions.request_url, and keep Socket Mode disabled. - The first deploy writes a bootstrap manifest and exits: This is expected when the Slack credentials do not exist yet. Create and install the app, add both credentials, then rerun the same command.
- MDA does not write the final manifest: Rerun
mda deploy . --configure-slackwithout--no-wait. The CLI needs the deployed Agent Server URL. - Slack cannot verify the request URL: Confirm the deployment is healthy, the URL on the app’s Event Subscriptions page matches
https://<agent-server>/channels/<name>/events, andSLACK_SIGNING_SECRETbelongs to that app. Redeploy after adding the Slack credentials, then apply the regenerated final manifest. - Mentions do not start runs: Subscribe to
app_mention, addapp_mentions:read, invite the bot to the conversation, and reinstall the app after changing scopes. - Direct messages do not start runs: Subscribe to
message.imand addim:history. - Thread replies do not start runs: Reply inside a thread where the agent previously participated. Subscribe to
message.channelsormessage.groups, add the matching history scope, and confirm the bot remains in the conversation. - The agent runs but does not reply: Confirm
auto_replyis enabled andSLACK_BOT_TOKENhaschat:write.
See also
- Channels overview: understand the provider-neutral channel model.
- Identity: configure authentication and caller ownership.
- Schedules: deliver scheduled results to Slack.
- Custom tools: attach a tool that uses
runtime.channel. - Deploy an agent: configure deployment secrets and inspect builds.
Connect these docs to Claude, VSCode, and more via MCP for real-time answers.

