OBTO

What OBTO MCP server does

The OBTO MCP server gives an AI client access to OBTO’s application-building and deployment platform. An agent can create a full-stack application, write its code, provision a database, and publish the application at a live obto.co URL. Workspaces are hosted by OBTO and are created when a user signs in.

The platform stores application code as database artifacts rather than as files in a local filesystem. Its supported artifact model includes pages, JavaScript modules, stylesheets, routes, and server scripts. The repository also provides instructions for the agent covering deployment order, backend service calls using the xe. pattern, and routing operations to the appropriate MCP tools.

The OBTO MCP server is intended for conversational application development. A user can describe an application, such as a landing page with a menu, gallery, and contact form, and have the connected AI client build and deploy it.

How it works

OBTO exposes one shared MCP endpoint:

https://app.obto.co/ms/mcp

OAuth identifies the signed-in user and their workspace. There is no personal endpoint and no API key to copy. After authentication, the AI client uses the MCP connection to perform platform operations during the conversation.

The supplied skill material is important because it teaches the agent OBTO-specific conventions. It explains how to represent code as database artifacts, which collections correspond to different application parts, how to deploy the backend before the frontend, and how to select MCP tools for particular operations. Without those instructions, an agent may generate code that does not follow the platform’s deployment conventions.

Setup and configuration

Create an account at obto.co/site/ob/signup using Google. The OBTO workspace is created automatically. Add the shared MCP endpoint to the client’s MCP configuration, then complete the OAuth approval when the client redirects you to sign in.

Claude Desktop can add the endpoint from its MCP server settings. Claude Web accepts the endpoint in a project. Cursor and VS Code require the server to be added through their settings.

For Claude Code or Claude Cowork, the repository provides a plugin marketplace entry:

/plugin marketplace add obto-inc/platform
/plugin install obto@obto

OpenAI Codex can install the corresponding plugin and finish authentication with codex mcp login obto. Other supported harnesses can use the repository’s SKILL.md or copy the skill folders into their skills directory.

Tools and capabilities

The OBTO MCP server supports workflows for:

• Creating full-stack application artifacts
• Writing pages, modules, stylesheets, routes, and server scripts
• Provisioning application databases
• Deploying backend and frontend components
• Uploading media and handling large deployment files through the provided skills
• Troubleshooting deployments and extending MCP usage through task-specific guidance

The repository packages seven task skills covering the build loop, deployments and large files, media uploads, memory, MCP extensions, troubleshooting, and platform boundaries.

Limitations and notes

OBTO does not use a local filesystem as the application’s source model; code is stored as database artifacts. Deployment order matters, with the backend deployed before the frontend. The platform-specific skill instructions are therefore part of the intended setup rather than optional generic documentation.

Authentication uses Google OAuth through the shared endpoint, and the README states that no API key is required. The material does not document environment variables, local deployment, or a list of individually named MCP operations. The OBTO MCP server is best suited to clients that can connect to a remote MCP endpoint and load the supplied skill or plugin instructions.