DEV Community

Runbear profile image Runbear Team
Runbear Team for Runbear

Posted on • Originally published at runbear.io

Build a Review-First Slack AI Agent from Claude Code with Runbear MCP

#mcp #agents #claude #ai

Runbear MCP lets you turn a Slack workflow into a reviewable AI teammate from Claude Code, without requiring a complete agent specification upfront.

Disclosure: We build Runbear. This guide describes the current early-access MCP server and links to the product documentation.

Safe to try: Adding Runbear MCP does not deploy anything to Slack. Agent changes and Slack deployment use separate permissions, and you can remove the MCP server at any time.

TL;DR

  • Add Runbear MCP to Claude Code with one command.
  • Authenticate through OAuth. No API key is pasted into chat.
  • Build and inspect a draft agent before connecting Slack.
  • Approve integrations, triggers, and deployment as separate steps.

Try it safely in Claude Code

Copy and run this command as one line:

claude mcp add --transport http runbear https://api.runbear.io/mcp/cc
Enter fullscreen mode Exit fullscreen mode

This only adds the MCP server to Claude Code. It does not install a Slack app, join a channel, create an agent, or send a message.

On the first Runbear operation, Claude Code opens a Runbear OAuth sign-in. Sign in to the Runbear organization you want Claude Code to manage; there is no long-lived API key to paste into the conversation.

Scope check: read:agents reads agent configuration, write:agents creates or updates agents, and deploy:agents is required to install an agent in Slack or join channels.

Then use a review-first prompt:

Create a draft agent called Meeting Actions.
Do not connect Slack, create a trigger, join a channel, or deploy anything.
First show me the proposed instructions, required integrations,
permissions, and a sample output.
Enter fullscreen mode Exit fullscreen mode

If the proposal looks right, connect only the apps you approve and request deployment in a separate turn.

Early-access note: Core agent creation, integrations, triggers, Slack deployment, and run inspection are available through MCP. Advanced model settings and knowledge-base synchronization still require the Runbear dashboard.

Claude Code creating a Runbear meeting-actions agent and connecting the required tools

One outcome-oriented prompt can produce a proposed agent configuration. Review it first, then approve integrations, triggers, and Slack deployment separately.

What is Runbear MCP?

Runbear MCP is a builder-style MCP server for Claude Code. It can create and change Runbear agents, connect capabilities, create jobs, deploy agents to Slack, and inspect recent runs.

The agent does not need to exist before the conversation starts. If you understand the operational problem but not every implementation detail, Claude Code can turn that incomplete description into a reviewable first version.

Start with the outcome, not the tool calls

You might begin with:

After a meeting, turn the notes into a concise list of owners and deadlines for the team.

That still leaves the source, trigger, safe actions, and destination unresolved. A practical building loop is:

  1. Describe the outcome. Start with the job, not internal tool names.
  2. Review the proposal. Check instructions, integrations, permissions, and sample output.
  3. Connect one capability at a time. Approve only the apps the workflow needs.
  4. Add the trigger and destination last. Keep Slack out until the draft is ready.

Keep creation separate from deployment

In practical terms:

  • write:agents can create or update an agent.
  • deploy:agents is required before an agent can be installed in Slack or join channels.

A draft workflow does not need a Slack destination. If a prompt explicitly names a channel such as #standup, deployment can happen in that turn. For a review-first workflow, leave the channel out until you approve the agent.

When the draft is ready, make the deployment step explicit:

The proposal looks good. Connect the approved apps and create the meeting trigger.
Before deploying, show me the exact Slack workspace, channel,
permissions, and sample message for final approval.
Enter fullscreen mode Exit fullscreen mode

Illustrative Slack output from the Runbear Meeting Actions agent

Illustrative result: after approval, the workflow posts owners and deadlines to the team in Slack.

When is this a good fit?

Runbear MCP is useful when:

  • The workflow is real but the complete agent specification is not.
  • Rules and capabilities will be decided incrementally.
  • You want a review step before the agent reaches Slack.
  • Developers and operators want to refine the same workflow from Claude Code.

It is less suitable if every advanced Runbear setting must be configurable entirely from the terminal today.

Current limitations

  • Advanced model settings and knowledge-base synchronization still use the Runbear dashboard.
  • Scheduled jobs currently post to Slack channels and DMs.
  • Skill-folder uploads use the separate skill-uploader plugin and require the Claude Agent SDK.

Ready to try it?

Add Runbear MCP to Claude Code:

claude mcp add --transport http runbear https://api.runbear.io/mcp/cc
Enter fullscreen mode Exit fullscreen mode

Start with the review-first prompt above. Nothing needs to reach Slack until you approve the destination.

To remove the server later:

claude mcp remove runbear
Enter fullscreen mode Exit fullscreen mode

Sources

Top comments (0)