Your agent joins
your team.
Connect any AI agent to your team's work cloud: the same skills your team writes, the same data they query, the same integrations and channels they use. One workspace. Every AI tool. A teammate's seat, not a guest pass.
Supported agents
runwork setup auto-detects and configures these agents.
ChatGPT
Light support · MCP-only
Codex
Full support
Claude Desktop
Full support
Claude Code
Full support
Cursor
Full support
Windsurf
Full support
Copilot (VSCode)
Full support
Copilot CLI
Full support
Gemini CLI
Full support
Antigravity
Full support
Full support means runwork setup wires up the desktop app, CLI, local agent config, skills, and MCP automatically. Light support connects over MCP only. Any MCP-compatible tool can also connect manually using your workspace endpoint URL and API key.
What's in your work cloud
Everything your agent gets access to the moment it connects.
Plus your team's own entities, workflows, agents, components, channels, and apps. The agent sees the same surface area your humans do.
How agents connect
MCP (Model Context Protocol)
Every workspace exposes an MCP server. Your AI agent connects once and gets tools for everything: entities, workflows, schedules, integrations, files, channels, skills.
Streamable HTTP and SSE transports. Standard protocol. Works with any MCP-compatible tool.
Learn moreRunwork CLI
Full terminal interface to the workspace. Every command supports --json for machine-readable output that AI agents parse natively.
runwork agents, runwork entities, runwork workflows, runwork integrations, runwork files, and more.
Learn moreSkills (SKILL.md)
Every app auto-generates a skill file with capability descriptions, API references, and domain knowledge. Your agent reads it and understands what's available.
YAML frontmatter + markdown. Portable to any AI tool. Downloaded from dashboard or synced via CLI.
Learn moreKnowledge files (CLAUDE.md, AGENTS.md)
When any AI agent creates or opens a Runwork app (via runwork init or runwork dev), the platform generates knowledge files on the fly. This works for any agent: Claude Cowork on the desktop creating a workflow, Claude Code building a full app, or Cursor adding an endpoint. The agent understands Runwork from the first prompt.
Created automatically. Always up to date. Works with Claude Cowork, Claude Code, Cursor, Codex, and any agent that reads project files.
Learn moreCommunity sync (skills.sh + MCP Registry)
Pull skills and MCP servers from public sources directly into your workspace. Your agent can install community-built tools from skills.sh + MCP Registry the same way it loads team-authored ones.
Discoverable from the marketplace, installable via CLI or dashboard, isolated per workspace.
Learn moreWhat your agent can do
Three modes once connected. Your agent talks with the team, operates the workspace, and builds new things.
Know
Work with full context of your business.
- Full workspace context
Your agent sees your data, connected integrations, and team skills from the first prompt, with no pasting background into every chat.
- Tap team skills
Use skills your teammates have built. What one person teaches their agent, every agent on the team can do.
- Hand off conversations
Pass an AI conversation to a teammate; they resume it natively in their own agent, with the full thread and context intact.
- Persistent memory
Conversations, prior apps you built together, and team context carry across sessions. Your agent picks up where it left off.
Do
Operate the workspace through the MCP server.
- Query and write entities
Read and update your team's data model: customers, projects, leads, vendors, anything you've defined.
- Call 3,200+ integrations
HubSpot, Stripe, Slack, Google, Notion, Linear. Every connected integration becomes a tool the agent can call directly.
- Trigger workflows & schedules
Run multi-step automations with parameters or kick off scheduled jobs on demand.
- Files & storage
Upload reports, download data, manage workspace buckets the same way humans do.
- Call public endpoints
Hit any HTTP endpoint your team has exposed from a workspace app: webhooks, APIs, internal tools.
Build
Create new apps, skills, and agents from conversation.
- Vibe-code full apps
Spin up a full-stack app from a prompt:
runwork init→runwork dev→runwork deploy. Frontend, backend, schema, integrations. - Author skills & agents
Write team-shared skills, register MCP servers, deploy task agents. Anything you build syncs to every teammate's tools.
- Drop in components
Reusable UI building blocks for the apps your agent assembles. No styling guesswork.
- Import 8,000+ workflow recipes
Convert any n8n workflow into a Runwork workflow in one command. Your agent picks the right recipe and wires it up.
- Delegate to agent sandboxes
Hand off heavy or long-running work to isolated compute environments with full code execution.
Connect in 60 seconds
That's it. Every connected agent now sees the same skills, data, integrations, and channels your team uses, and stays in sync as the workspace evolves.
Setting up Runwork for non-technical teammates? Send them to the Desktop app instead. Same `runwork setup`, guided UI.