Open-source Control Plane
for your AI Agents.

One MCP endpoint for all your agents, tools, and models. Install locally, connect from any browser, track every token and dollar in real time.

Star on GitHub

~/studio

$ bunx decocms

deco studio v0.4.2

✓ Running at http://localhost:3000

CONNECTIONSSTATUSLATENCY

postgres-mcp● active12ms

github-mcp● active89ms

openai-mcp● active340ms

seo-toolkit● active52ms

4 connections active · postgres, github, openai, seo-toolkit

AGENTSSTATUSTASK

Researcher● doneCatalog schema mapped

Writer● doneBlog draft generated

SEO Agent● runningAdding meta tags...

3 agents hired · 2 done, 1 running, 1 queued

Cost: $0.04Tokens: 12,847Duration: 8.2s

How it works

The operating system for AI teams.

Hire specialized agents, give them access to your tools, and let them deliver real projects — with planning, verification, and observability built in.

Hire Agents

Browse and hire specialized AI agents with specific skills. Each agent knows how to use specific tools and comes with battle-tested prompts. Like hiring contractors — but they start immediately and scale infinitely.

Agent marketplace with verified skills

Custom agents from your own prompts

Agents compose — they can call each other

Usage tracking and cost attribution per agent

Connect Tools

Set up your MCP connections through a web UI with one-click OAuth. No JSON configs. Monitor latency, errors, and costs per connection.

GitHub, Slack, Postgres, and 50+ integrations

One-click OAuth authentication

Per-connection cost tracking — see spend on OpenRouter, Perplexity, Firecrawl

Granular RBAC — share tools, not credentials

Work Anywhere

Studio runs locally but you control it from any browser. See what your agents are doing, approve actions, check costs, and start new tasks — from your laptop, phone, or tablet.

Web dashboard for all your local agents and tools

Real-time monitoring — see tool calls and costs live

Approve, pause, or redirect agents on the go

Declarative planning — define outcomes, agents deliver

See it in action

Declare what done looks like. Studio derives the work, assigns agents, and verifies the outcome. Here's a real workflow:

D-01Declaration

"Blog posts are generated from product data and published with SEO metadata."

M-01M-01: Blog generation pipeline produces draft posts from catalog

Researcher

Analyze product catalog schema

postgres-mcpDone

Writer

Generate blog post from product data

openai-mcpDone

SEO Agent

Add meta tags, structured data, and alt text

seo-toolkit-mcpRunning

Publisher

Create PR with blog post and images

github-mcpPending

VerificationVERIFIED — Blog post exists, SEO metadata present, PR created with passing CI.

Observability

Know what every agent costs.

See token spend per connection — OpenRouter, Perplexity, Firecrawl, all of it. Latency, errors, bottlenecks. One dashboard.

Cost attribution

OpenRouter: $4.23 today

Per-connection spend tracking across all providers

Latency monitoring

p95: 340ms GitHub MCP

Real-time latency percentiles for every connection

Error rates

Sentry MCP: 2 / 1,847 calls

Errors surfaced instantly — not buried in logs

Scale

From your laptop to your org.

Same mental model everywhere: your agents, your skills, your connections, your projects. Start local. Go cloud. Scale to teams.

Local

bunx decocms on your laptop. SQLite. Private.

Cloud

Team

Invite people. Roles. Shared connections. Wallet-based billing.

Enterprise

Self-hosted. Your infra. Your rules.

Why deco Studio?

An honest comparison. We built Studio because nothing else gave us what we needed.

Feature	deco Studio	Raw MCP setup	DIY
Open source	Yes, fully	Depends	Depends
Any model	Any provider + local	Manual setup	Manual setup
Tool management	Centralized MCP mesh	Per-project JSON configs	Scattered configs
Agent marketplace	Browse, hire, compose	N/A	Build everything
Project planning	Declarative engine	N/A	Manual / Jira
Observability & costs	Built-in dashboard	None	Add-on required
Remote access	Browser + mobile	N/A	Custom SSH/VPN

FAQ

Frequently Asked Questions

Is this a Claude Code alternative?

What's the difference between local Studio and studio.decocms.com?

Is this really free and open source?

How is this different from MCP Mesh?

Can I use it with Claude, GPT, and local models?

What does 'remote control' mean?

Do I need to use Deco's cloud?

What's declarative planning?

Open-source Control Planefor your AI Agents.