Creator OS
ai-agentsmcpself-hostedslacksocial-media

Self-Hosted Slack With AI Agents: 2026 Social Team Guide

Self-hosted Slack with AI agents lets your social team run replies, monitoring and campaigns in one workspace without per-seat SaaS. Here is how to wire it up.

Creator OS · November 12, 2026 · Updated November 17, 2026 · 10 min read

Self-hosted Slack with AI agents means you run the chat app on your own server and AI agents sit in the same channels and DMs as your team. agentes-chat is an open-source project that does this: agents run on Claude, ChatGPT/Codex, Gemini or Grok, with MCP support, so replies, monitoring and campaign talk happen in one workspace instead of five tabs.

  YOUR SERVER (self-hosted)
  +---------------------------------------------+
  |  #campaigns   #replies   #alerts    DMs     |
  |     |            |          |        |      |
  |  humans       humans     humans   humans    |
  |     \            |          |        /      |
  |      +-----------+----------+-------+       |
  |                  |                          |
  |           AI agent (Claude, ChatGPT,        |
  |           Gemini, Grok) via MCP             |
  +------------------+--------------------------+
                     |  MCP over HTTPS
                     v
        Creator OS hosted MCP server
        https://mcp.creatoros.ca/mcp
                     |
        posts, comments, DMs, analytics,
        Skool, Reddit, Discord, WhatsApp
Agents and humans share one chat workspace; the agent reaches your social accounts through MCP.

What self-hosted Slack with AI agents actually means for social teams

You own the chat server, and the agent is a member of the team, not a separate bot dashboard.

Normal setup: your social team talks in Slack, your scheduler lives in another tab, your replies inbox in a third, and someone copies links between them. A self-hosted chat app with agents collapses that. The channel where you argue about a hook is the channel where the agent schedules the post. The DM where a client asks for a stat is the DM where the agent pulls it.

Two things make that possible:

  • The agent can read context. It sees the thread where you decided the angle, so it does not need a fresh brief for every task.
  • The agent can act. Through MCP it reaches real accounts, not just a chat window.

Self-hosted is the deployment model. You run the server, you hold the database, you control who joins. That is the part that matters for privacy, and it is separate from which model the agent uses.

If you want the background on agents that post, reply and report on their own, read Social Agents open source and watch the walkthroughs on the KevBuildsApps YouTube channel.

Get started for $0

How agentes-chat runs AI agents in Slack channels and DMs

The agent joins channels like a teammate and answers when addressed, with the same rules for DMs.

agentes-chat is an open-source Slack alternative you host yourself. Teams and their AI agents share channels and DMs in one place. The agents can run on Claude, ChatGPT/Codex, Gemini or Grok, and the project supports MCP, which is the hook that matters for a social team.

The practical shape for a creator business:

  1. #replies is where the agent drops comments that need a human eye.
  2. #campaigns is where you brief a launch and the agent drafts copy for each platform.
  3. DMs are where a client or collaborator asks for numbers and gets them without opening a dashboard.

You do not need to rebuild your whole stack to test this. Spin up the self-hosted app, connect one MCP server, and give the agent one job: read comments and post the ones that need approval into a channel. If that earns its keep, add scheduling.

For the wider pattern of agents that take over a social workflow end to end, see a fleet of AI agents for your content workflow.

Which models and MCP support matter for a social team setup

Model choice matters less than whether the agent can call tools safely and which workspace it can touch.

Layer What to decide Why it matters for social
Chat app Self-hosted (agentes-chat) or hosted Where humans and agents talk
Model Claude, ChatGPT/Codex, Gemini or Grok Writing quality, cost, latency
Tool access MCP server for your social accounts Lets the agent actually post and reply
Permissions Read-only or read and write Stops a wrong reply before it ships
Scope One workspace per connection Keeps clients apart

Creator OS ships a hosted MCP server at https://mcp.creatoros.ca/mcp. Nothing to install. You add it in the agent’s tool settings, sign in to Creator OS, and pick the workspace. Each connection gets its own key, listed under Settings > Connected apps. On agency plans, one client per connection.

Permissions are the part people skip. You can create a read-only connection, and Creator OS refuses writes from it. An agent that can only read is a good first step: it watches comments and analytics and reports into the channel. Destructive tools (delete post, delete comment, delete automation, disconnect account and so on) are marked so your AI app asks before running them.

The tool surface is about 160 tools across posts, analytics, comments, DMs, automations, webhooks, Skool, Reddit, Discord, WhatsApp, broadcasts, Slack, Shopify, the blog, trackable links, skills and the Ads add-on. Names you will see include create_post, list_comments, reply_to_comment, send_message, get_post_analytics, get_best_time_to_post and links_create. Full reference at the MCP docs.

Can you run community management without per-seat SaaS costs

Yes, and the savings come from two places: seats and the tool count.

Seat-based SaaS charges per person. A six person social team on a per-seat scheduler pays six times. A self-hosted chat app plus an MCP connection does not bill per human. Your agents are not seats either.

The second saving is consolidation. If your chat app, your scheduling tool, your reply inbox and your reporting tool are four subscriptions, the agent removes the need for three of them because it does the work through one MCP server.

On the Creator OS side, the numbers are flat:

  • Creator Premium: $14.99/month or $59.99/year for one set of socials, 11 platforms plus Skool, one account on each. Includes the iOS app.
  • Agency 5: $49.99/month or $399/year for five client workspaces. Unlimited team members.
  • Agency 10: $99/month or $599/year for ten client workspaces.

Agency plans include unlimited posts, comments and messages, unlimited team members, and a scoped API key per client workspace. API keys, the MCP server, the CLI and the Creator OS agent skills are included in every plan. The chat app itself is open source and you host it. Creator OS plans start at $14.99/month for 11 platforms plus Skool, with the API, the MCP server and the CLI included.

What you do pay for: your server, and whatever the model costs. Creator OS does not ship its own model, so you bring the one you already use.

Self-hosting as a privacy option for creator businesses

Self-hosting keeps message history, member lists and client conversations on hardware you control.

For a creator business with client accounts, that is the real argument. A hosted chat app stores your DMs and channel history on someone else’s servers. A self-hosted app stores them on yours. Delete a workspace, and the data goes where you say.

Two things to keep straight:

  1. Self-hosting the chat app does not make your social posts private. They still go out on Instagram, TikTok, YouTube or wherever you publish. What self-hosting protects is the conversation around the work.
  2. MCP connections are scoped. A Creator OS API key is pinned to one workspace, so an agent holding one workspace’s key cannot post to another. That holds whether the agent runs on your server or in a hosted AI app.

If your team handles DMs from customers, read the authentication docs before you hand an agent write access, and start it read-only until you trust the output.

AI agents in Slack vs standalone social agents: what changes for your workflow

A standalone social agent works in its own window; a chat-native agent works in the room where the decisions happen.

Creator OS supports both. The open-source Social Agents harness runs your socials locally with its own dashboard. Start it with npm start creatoros social-agents. It interviews you about your brand, then posts, replies to comments and DMs, runs automations and reports. You bring your own model: a logged-in Claude Code session, an ANTHROPIC_API_KEY, or any Anthropic-compatible API.

That harness is great for a solo creator. A team that already lives in channels gets more from the chat-native version, because the approval step is a message instead of a tab switch.

Standalone agent Chat-native agent
Where you work Its dashboard Channels and DMs
Approval In the tool A reply in the thread
Best for Solo creator Team with clients
Setup Local harness Self-hosted chat plus MCP

If you want the code-editor version of the same idea, the self-hosted remote control for coding agents post covers the pattern.

How to wire an agent into monitoring, replies and campaign coordination

Start with monitoring, add replies, then campaigns. Each step reuses the same MCP connection.

  1. Connect MCP read-only. Add https://mcp.creatoros.ca/mcp in your agent’s tool settings, sign in, pick the workspace, choose read only. The API refuses writes from that connection.
  2. Post a daily digest. Ask the agent to run get_daily_metrics, get_follower_stats and list_comments, then summarize into #alerts with the three comments that need a human answer.
  3. Add replies with a guard. Switch to a read and write connection. Keep the agent drafting replies into the channel and only sending after a human says go. Destructive tools ask first on their own.
  4. Coordinate a campaign. In #campaigns, ask for a launch plan across your connected platforms. The agent calls check_caption_length per platform and create_post with schedule_at and a timezone. Networks you have not connected come back in missing_platforms; the rest still go out.
  5. Close the loop. After publish, ask for get_post_analytics and get_best_time_to_post, then feed the numbers into the next brief.

For the command-line version of steps 2 to 5, the CLI does the same work. A quick check after posting:

creatoros analytics:posts --platform instagram --sortBy engagement --limit 10
creatoros accounts:health
creatoros inbox:comments --platform instagram --limit 20

If you are wiring this inside Claude Code, the plugin adds the server, campaign builder skills and a publish guard that asks Proceed or Cancel before anything is published. Install with /plugin marketplace add kevinbadi/creatoros-claude-plugin, then /plugin install creatoros@creatoros. It needs Claude Code 2.1.287 or later. More at the Claude Code docs. The launch video shows the plugin in action.

FAQ

What is self-hosted Slack with AI agents?

It is a chat app you run on your own server where AI agents are members of channels and DMs, alongside your team. agentes-chat is an open-source example: agents run on Claude, ChatGPT/Codex, Gemini or Grok, with MCP support. You control the server and the data, and the agent acts inside the same conversation where your team works.

How do AI agents handle social replies inside Slack channels?

The agent reads the comment through a tool, drafts a reply, and posts it into a channel for approval before it sends. In Creator OS, the tools are list_comments and reply_to_comment, reached over MCP, and a read-only connection blocks writes until you switch permissions. You can also have the agent file urgent comments into #alerts and leave the rest for a daily pass.

Is self-hosted Slack with AI agents more private than SaaS?

It is more private in one specific way: message history, member lists and client conversations sit on hardware you control rather than a vendor’s. It does not make your published social posts private, since those still go out on the public platforms. Keep API keys scoped per workspace and start agents read-only.

Can I connect MCP tools to a self-hosted Slack agent?

Yes. agentes-chat supports MCP, so you can point the agent at a hosted MCP server such as https://mcp.creatoros.ca/mcp. Sign in to Creator OS, pick the workspace, and choose read only or read and write. Each connection gets its own key, listed under Settings > Connected apps.

How much does it cost to run AI agents in Slack without per-seat pricing?

Creator OS does not charge per seat. Creator Premium is $14.99/month or $59.99/year, and agency plans include unlimited team members across five or ten client workspaces. You also pay for your own server and for whatever model the agent uses, since Creator OS does not ship its own model.

Get started with Creator OS

Sign up at https://www.creatoros.ca/sign-up. Creator Premium is $14.99/month for 11 platforms plus Skool, with the API, the MCP server and the CLI included, and new subscribers start for $0 with a free trial.

Then add https://mcp.creatoros.ca/mcp to your agent, connect it read-only, and give it one job for a week. The docs cover the rest, and the KevBuildsApps YouTube channel has the walkthroughs.

Get started for $0
Free trial for new subscribers. Cancel anytime.

Keep reading