What is the Matagi MCP server?

It is a single remote MCP endpoint that gives any MCP-capable client — Claude, Cursor, Codex, ChatGPT — the ability to provision real infrastructure and operate Matagi agents on your behalf. Anything you could do in the Matagi dashboard becomes a tool your AI can call.

Do I need an API key?

No. The server uses OAuth 2.1 with Dynamic Client Registration, so your client registers itself and signs you in through a browser consent screen. There are no API keys to copy, paste, or rotate.

Do I need an account first?

Not separately. The first time you connect, the consent screen signs you in and headlessly creates your Matagi workspace with trial credit. You can also sign up ahead of time at app.matagi.ai.

Which clients are supported?

Anything that speaks MCP over Streamable HTTP: Claude Code, Claude Desktop, Claude.ai on the web, Cursor, OpenAI Codex, ChatGPT, VS Code, and more. One connection works across all of them.

Where do my credentials and secrets live?

You authorize each external tool once inside Matagi. Access is granted to your resources through the Matagi proxy, so your real API keys never appear in generated code — everything is proxied and encrypted. You can revoke any access at any time with a full audit trail.

How much does it cost?

Connecting and signing up is free with a trial, no card required. Agent runtime, model calls, and infrastructure are billed at cost with no markup. See matagi.ai/#pricing for plans.

MCP Server | Matagi

/01 · Overview

Your AI, with hands.

Most coding agents can write a plan but can't run it — someone still has to stand up the servers, wire the databases, and manage the credentials. The Matagi MCP server closes that gap. It exposes everything Matagi can do as MCP tools, so the model goes straight from instruction to working agent with you out of the middle.

As your AI builds, Matagi auto-provisions whatever it needs and wires it together: infrastructure, memory, integrations, and model calls. You connect the server once and switch freely between clients — the same workspace, agents, and resources follow you everywhere.

/02 · What you can build

Everything, from one connection.

Provision infrastructure

Serverless functions, batch jobs over CSVs, GitHub repos, Vercel deployments, and isolated Postgres databases — minted on demand and tracked per project.

Model & tool proxy

Anything you build calls every major model (Claude, GPT, Gemini) and 3,000+ tools through the Matagi proxy. Credentials are injected server-side — never in the generated code.

Run autonomous agents

Create, spawn, seed, and chat with always-on AI workers. Wire up Slack, Telegram, Discord, email, or web chat, and manage their files and memory.

Billing & workspace

Read your plan, balance, and usage, mint top-up links, and manage agent slots — all without leaving your editor.

/03 · Server endpoint

One URL, every client.

The Matagi MCP server is a single remote endpoint served over Streamable HTTP. Point any MCP client at it — there is no local process to install and nothing to self-host.

Production endpoint

https://mcp.matagi.ai/mcp

Transport

Streamable HTTP

Auth

OAuth 2.1 + PKCE

Registration

Automatic (DCR)

/04 · Authentication

No keys. Just sign in.

Authentication is standard OAuth 2.1 with PKCE and Dynamic Client Registration (RFC 7591). Your client discovers the server's OAuth metadata, registers itself, and walks you through a browser consent screen — no API keys, no app registration, nothing to copy by hand.

Add the server URL to your client and trigger the first tool call (or the explicit authenticate action).
A Matagi consent screen opens in your browser. Sign in with your work email.
Approve access. Matagi creates your workspace with trial credit if you're new, then hands the client a token.
You're connected. Tokens refresh automatically — you won't be asked again unless you revoke access.

/05 · Connect your client

Pick your client.

Choose where you work. Each connects to the same https://mcp.matagi.ai/mcp endpoint and the same workspace.

Connect Claude Code

Add the server from your terminal, then authenticate in-session.

$ claude mcp add --transport http matagi https://mcp.matagi.ai/mcp

Run /mcp inside Claude Code and select matagi to open the consent screen. Approve it once and the tools are live.

/06 · How it works

What happens after you connect.

When your AI calls a Matagi tool, the server provisions cloud resources in isolated, per-workspace sandboxes and tracks each one in a registry addressed by an opaque id. Resources are owned by your workspace, so every member of your team can operate them.

Whatever gets built can reach the outside world through the Matagi proxy: a model proxy for native access to every major LLM, and a connection proxy that forwards authenticated calls to 3,000+ tools. The OAuth tokens and API keys for those tools are injected at runtime and never touch the generated code or the model context.

The result is a clean separation: the model decides what to build, Matagi runs it safely, and your credentials stay proxied, encrypted, and revocable with a full audit trail.

/07 · FAQ

Common questions.

Connect it and start building.

Add the endpoint to your client, sign in once, and tell your AI what you want. Matagi provisions the rest.

Connect your client Create an account