Runbear Skills

Skills for exposing Claude Agent via HTTP, Slack, and Discord.

Table of Contents

• Use Cases
• Installation
• dispatch-http
• dispatch-slack
• dispatch-discord
• skill-uploader
• runbear
• License

Use Cases

• Introduce a Claude Code project to team members — Let your team interact with a project through Slack or Discord without needing Claude Code installed locally
• Interact with a Claude Code project running on a Mac Mini — Access a headless Claude Code instance remotely via HTTP, Slack, or Discord
• Query a service that uses Claude Code as backend — Expose Claude Code as a REST API for other services to consume
• Use Siri to command Claude Code — Chain Siri Shortcuts with the HTTP API to control Claude Code by voice

Installation

/plugin marketplace add runbear-io/skills
/plugin install dispatch-http@runbear-skills
/plugin install dispatch-slack@runbear-skills
/plugin install dispatch-discord@runbear-skills
/plugin install skill-uploader@runbear-skills
/plugin install runbear@runbear-skills

dispatch-http

Expose Claude Code as REST API endpoints.

Skill	Description
dispatch	Start an Express server that exposes Claude Code/Cowork via REST endpoints
expose	Expose the local dispatch-http server to the internet via a Cloudflare quick tunnel

Usage

/dispatch-http:dispatch
/dispatch-http:dispatch port 8080
/dispatch-http:expose

dispatch-slack

Connect Claude Code to Slack via Socket Mode. Supports streaming responses.

Skill	Description
dispatch	Start/stop the Slack bot server that connects Claude Code to Slack
init	Create and configure a Slack bot app with OAuth token rotation

Usage

/dispatch-slack:init
/dispatch-slack:dispatch [start|stop|status] [/path/to/project]

Running modes

The dispatch skill supports two modes:

• Local — Run directly on the host machine. Uses local Claude Code authentication.
• Docker — Run in a container for sandboxed filesystem access. Requires CLAUDE_CODE_OAUTH_TOKEN (generate with claude setup-token) or ANTHROPIC_API_KEY in .env.

dispatch-discord

Connect Claude Code to Discord. Supports streaming responses via progressive message editing.

Skill	Description
dispatch	Start/stop the Discord bot server that connects Claude Code to Discord
init	Create and configure a Discord bot via the Developer Portal

Usage

/dispatch-discord:init
/dispatch-discord:dispatch [start|stop|status] [/path/to/project]

Running modes

The dispatch skill supports two modes:

• Local — Run directly on the host machine. Uses local Claude Code authentication.
• Docker — Run in a container for sandboxed filesystem access. Requires CLAUDE_CODE_OAUTH_TOKEN (generate with claude setup-token) or ANTHROPIC_API_KEY in .env.

skill-uploader

Deploy local Claude skill folder(s) to a hosted Runbear Agent SDK agent. Reads a local skill folder, runs a safety preflight, and deploys each SKILL.md through the Runbear upload_local_skill_to_agent MCP tool.

Skill	Description
upload	Upload local skill folder(s) to a Runbear agent identified by an app URL or app UUID

Prerequisites

• The Runbear management MCP server must be connected in your Claude Code session — it provides the upload_local_skill_to_agent tool.
• The target Runbear Agent SDK app URL or app UUID (--agent).

Usage

/skill-uploader:upload ./my-skill --agent https://app.runbear.io/agents/<appId>
/skill-uploader:upload ./my-skill --agent <appId>
/skill-uploader:upload ./my-skill --agent <appId> --overwrite
/skill-uploader:upload ./internal-skills --agent <appId>

The backend deploys one skill per call. If the folder contains multiple SKILL.md files, each is uploaded independently and reported per skill.

runbear

Deploy a whole local Claude Code project to a hosted Runbear Agent SDK agent and configure the MCPs and skills it needs, so colleagues can use your agent without you doing it on their behalf. The hosted agent runs against the same CLAUDE.md, skills, subagents, docs, and code you have locally, with the directory layout preserved.

Skill	Description
deploy	Deploy a local project's files, select which local MCPs the agent gets (shared vs per-user auth per MCP), and curate its skills — all saved to a committed manifest for one-command redeploys
connect-slack	Connect a Runbear Agent SDK agent to Slack — creates a workspace-specific custom bot by default, and can join channels

What deploy configures

• Files — uploaded via a signed URL (bytes never pass through the model context).
• MCPs — pick from your locally-configured MCP servers (project + user scope); each is attached to the agent as a Runbear catalog app (attach_app) or a custom MCP (attach_custom_mcp). Per MCP you choose shared auth (one org credential, userType: app) or per-user auth (each colleague signs in as themselves, userType: user — the default). OAuth MCPs attach cleanly; you're prompted before any local secret is read to vault a static-auth MCP.
• Skills — curated from your project (.claude/skills/) and your user scope (~/.claude/skills/); the chosen set ships in the upload.
• AI suggestions — the skill reads your CLAUDE.md and pre-selects a recommended MCP/skill set you can accept or edit.
• Manifest & redeploy — choices are written to a committed .runbear/deploy.json. Re-running deploy reproduces the exact configuration with no re-prompting; --add-mcp / --remove-mcp / --add-skill / --remove-skill adjust it.

Limitations: local stdio MCPs (npx/uvx/command servers) can only reach the hosted agent through a Runbear catalog equivalent — with no match they're reported as un-deployable. Vaulting a static-auth custom MCP requires reading its secret into context, which only happens after you confirm a per-MCP prompt.

Prerequisites

• The Runbear management MCP server must be connected in your Claude Code session — it provides create_project_upload, finalize_project_upload, search_apps, attach_app, attach_custom_mcp, list_agent_tools, and detach_tool (plus create_agent for the --new flow).
• The claude CLI on PATH — used to enumerate local MCPs (claude mcp list / claude mcp get).
• The target Runbear Agent SDK app URL, app UUID, or agent name (the first argument) — a name is fuzzy-matched, with an interactive picker when more than one agent matches — or --new "<agentName>" to create a fresh Claude Agent SDK agent and deploy to it. An existing target agent must be of type Claude Agent SDK. Resolving a name uses the list_agents tool.
• Local tools: git, zip, and curl on PATH.

Usage

Deploy to an existing agent:

/runbear:deploy <appId>
/runbear:deploy https://app.runbear.io/agents/<appId>
/runbear:deploy "My Agent"
/runbear:deploy <appId> --cwd ./my-project
/runbear:deploy "My Agent" --cwd ./my-project --overwrite

Or create a new Claude Agent SDK agent and deploy to it in one step:

/runbear:deploy --new "My Agent"
/runbear:deploy --new "My Agent" --cwd ./my-project

Redeploy with the saved settings (no re-prompting), optionally editing the set:

/runbear:deploy "My Agent"
/runbear:deploy "My Agent" --add-mcp linear --remove-skill .claude/skills/old

The backend caps a deploy at 500 files / 25 MiB decompressed / 50 MiB zip, and blocks secrets, .env, .git/, .runbear/, node_modules/, .mcp.json, and Claude settings files. Use --overwrite to replace files that already exist in the agent workspace. .runbear/deploy.json (the deploy manifest) is committed to your repo but excluded from the upload.

Connect an agent to Slack (a workspace-specific custom bot by default):

/runbear:connect-slack "My Agent"
/runbear:connect-slack <appId> --bot-name "My Bot" --channel project-team
/runbear:connect-slack <appId> --mode default

Custom mode returns a Runbear web-app link to finish creating the Slack app; joining public channels is then handled through the MCP. Use --mode default to attach the shared @Runbear bot instead.

License

MIT