Key concepts

How Studio is organized

Studio uses a small set of concepts that build on each other. Once you understand them, everything else follows.

• Organization — your top-level workspace. Everything belongs to an org: members, agents, connections, automations, and logs.
• Agent — an AI configured for a specific job: a focused set of tools, clear instructions, and a narrow purpose. Agents are the working surface of Studio. See Agents.
• Connection — a tool or service you’ve plugged in (GitHub, Slack, a database, your own API). Credentials are stored once and shared securely. See Connections.
• Thread — a conversation session inside an agent. Studio saves the history so you can come back to ongoing work.
• Automation — an agent run that fires on a schedule or in response to an event. See Automations.
• AI Provider — a configured AI model (Anthropic, Google, OpenRouter, or the Deco AI Gateway) that powers your agents and decopilot. See AI Providers.
• Member — a person in your organization with a role and permissions.
• Project — a planned second level of hierarchy that will group agents, connections, and context around a shared goal or team. See Projects.

Anatomy of the Studio UI

The agent shell

This is where day-to-day work happens. The persistent sidebar on the left shows your organization’s agents. Click an agent and you get:

• Threads list (left) — every conversation you’ve had with this agent.
• Chat (center / bottom) — where you talk to the agent.
• Agent panel (right) — the agent’s configuration: its attached connections, its instructions, and tabs for Settings, Automations, and (when linked to GitHub) Git.

Everything you need to use, configure, and monitor a single agent lives on this one screen.

Settings

Everything that isn’t tied to a single agent lives under Settings. Open it from the bottom-left of the sidebar. Inside you’ll find:

• Organization — General, Brand Context, AI Providers
• Build — Connections, Agents (the catalog), Automations, Store
• Manage — Monitor, Members, Roles, Security (SSO)
• Extensions — Plugins
• Account — Profile & Preferences

When other pages in these docs say “go to Connections” or “open Monitor,” they mean these Settings sections.

How the protocol works

deco Studio is built on the Model Context Protocol (MCP), an open standard for connecting AI clients to tools and services. You don’t need to understand MCP to use Studio, but knowing the three building blocks helps:

• Tools — actions an AI can take (search a database, send an email, create a file)
• Resources — read-only data sources that give an AI context (documents, configs, org data)
• Prompts — reusable templates for common workflows, with variables you can fill in

Studio sits between your AI clients (like Claude Desktop or Cursor) and your tools, handling authentication, routing, and logging for every request.

How a request flows through Studio

When an agent calls a tool, here’s what happens:

 ┌─────────┐                    ┌──────────────┐                  ┌────────────┐
│ Client  │ ──[authenticate]──▶│  deco Studio │ ──[with creds]──▶│ MCP Server │
│ (Cursor)│                    │              │                  │  (GitHub)  │
└─────────┘ ◀──[response]──────└──────────────┘ ◀─[response]─────└────────────┘
                                      │
                                      └──▶ [audit log] 

1. The client authenticates to Studio
2. Studio checks whether the client has permission to use the requested tool
3. Studio retrieves the stored credentials for that connection
4. Studio forwards the request to the upstream service
5. Studio logs the invocation — tool, timing, inputs, outputs, and who called it
6. Studio returns the response to the client

Credentials never leave Studio, every action is logged, and access is consistent regardless of which client made the request.

Why monitoring is built in

Without a control plane, every client manages its own logs — or doesn’t log at all. Studio logs every tool call that flows through the platform, creating a single place to debug failures, watch performance, and audit who accessed what. There’s no extra setup. See Monitoring.