Set it up from your coding client. Runbear keeps it running.

Install the Runbear MCP server once, then describe the workflow in Claude Code, Codex, or Cursor. Runbear drafts the trigger, the apps, and the output for your review. After you deploy it, it runs for your team.

OAuth in your browser

• Three coding clients
• Review before deployment

Claude Code example showing Runbear drafting an agent

When a Zoom meeting ends, wait 5 minutes, pull the notes from Granola, post the action items to #product-sync, open a Linear ticket for each one, and mention the owner.

• Trigger: Zoom meeting ends, then 5 min
• Apps: Granola · Slack · Linear
• Output: Slack recap and assigned Linear tickets

Drafted→You review→Runs in Runbear, every time

Connect Runbear to your coding client

Choose the client you already use, run its install step, and complete OAuth in the browser.

• Claude Code
• Codex
• Cursor

Claude Code

Requires remote HTTP MCP support.

Install command

claude mcp add --transport http runbear https://api.runbear.io/mcp/cc

1. Connect
   Run the command in the project where you want Claude Code to use Runbear.

2. Authenticate
   Run /mcp, select Runbear, and complete OAuth in the browser. Runbear uses your only or saved default organization.

3. Verify
   /mcp should show Runbear as connected. You can also inspect the server from your terminal.

Verification commands

claude mcp list
claude mcp get runbear

OAuth without pasted API tokens

The first Runbear action opens a browser window. Sign in and return to your coding client. Runbear uses your only organization or saved MCP default, and you never paste a Runbear API token into chat.

Tied to your Runbear account
Bound to your only or saved default organization
No API token pasted into chat

Create a review-first agent

The starter prompt asks your coding client to draft the instructions and propose the app connections, schedule, and output. It shows the complete configuration without connecting apps, creating triggers, or deploying.

First agent prompt

Create a new Runbear agent named Weekly Standup.
Draft instructions that collect team updates and summarize blockers.
Propose the app connections, weekly schedule, and output this agent would need.
After creating it, show me the complete configuration and recommendations.
Do not connect apps, create triggers, or deploy it yet.

Already have a Claude Code project?

Runbear Agent Skills deploys an existing Claude Code project into a hosted Agent SDK agent. It does not connect that agent to Slack automatically.

Describe work your team already does

Runbear turns the request into a draft blueprint: when the agent should run, which apps it needs, and what it should produce. You review each decision before connecting or deploying it.

Meeting follow-up

Turn every meeting into assigned work nobody has to write up.

• Trigger: Zoom meeting ends, then 5 min
• Apps: Granola · Slack · Linear
• Output: Slack recap and assigned Linear tickets

Support triage

Turn new support conversations into a reviewable action plan.

• Trigger: New Zendesk ticket
• Apps: Zendesk · Slack
• Output: Triage summary and draft reply

Operations briefing

Give operators the context they need before deciding what to do.

• Trigger: Weekdays at 9:00
• Apps: Notion · Slack
• Output: Team briefing with open decisions

What the first draft should show

• Runbear appears connected in /mcp
• OAuth returns you to Claude Code
• The response presents Weekly Standup as undeployed
• Claude Code shows the draft instructions and complete configuration
• Proposed app connections, weekly schedule, and output are ready to review

You decide what goes live

1. Review the instructions and recommendations
2. Connect an approved app
3. Create the approved trigger or schedule
4. Run a test
5. Ask the client to deploy when ready

Turn the draft into a working agent

When the blueprint looks right, use the MCP reference to connect apps, create triggers, test the workflow, and deploy. It also documents permissions, traces, security, and limitations.