decocms/studio

What it solves

deco Studio provides a secure, centralized control plane for organizations to deploy and manage AI agents. It solves the problem of "fragmented AI rollouts" by consolidating model routing, tool authentication (via MCP), access control, and cost tracking into a single platform, preventing the need to share raw credentials or rebuild infrastructure for every new team or agent.

How it works

Studio acts as a governance layer between AI clients (like Cursor or Claude Desktop) and the underlying models and tools. It uses the Model Context Protocol (MCP) to create a single endpoint that routes requests to various agents. These agents package specific instructions, tools, and model policies.

Key technical components include:

• Encrypted Vault: Stores system tokens (GitHub, Slack, etc.) so agents can access tools without exposing credentials to users.
• Virtual MCPs: Allows the system to dynamically filter or sandbox which tools are exposed to a specific agent.
• Model Router: Makes the AI layer interchangeable, supporting OpenRouter or direct providers (OpenAI, Anthropic, etc.).
• Observability Pipeline: Uses OpenTelemetry and ClickHouse to trace every call, calculating latency and cost per user or agent.

Who it’s for

It is designed for organizations and engineering teams that want to move beyond individual AI assistants to a governed, team-wide AI infrastructure with SSO, role-based access control (RBAC), and audit logs.

Highlights

• Centralized MCP Management: Connect private systems once and share capabilities across the org via a single endpoint.
• Interchangeable Models: Switch between providers per agent or tool without changing the client configuration.
• Enterprise Governance: Built-in SSO, RBAC, and detailed cost attribution/audit logs.
• Adaptive UI: Project-based interfaces that change based on the agents and connections assigned to that project.
• Flexible Deployment: Can be run locally with embedded PostgreSQL or deployed via Docker and Kubernetes (Helm).