What it does

Brod connects to an AI agent as a remote MCP server at https://brod.sh/mcp. In Claude Code, the connection is added with the command claude mcp add --transport http brod https://brod.sh/mcp. Brod also lists connection steps for the Claude web and desktop app, Cursor and VS Code. Any other agent that speaks MCP can connect over HTTP. A bearer token from the workspace's Connections page can be used when the agent has no browser sign-in. The user then asks the agent to put the tool online with Brod. The agent reads the workspace's groups and roles with get_workspace. It then deploys the app with deploy_app. The app goes live on HTTPS at its own address. It stays private to its owner until a person confirms sharing. An app is a folder. Server code is a single _worker.js file. Static files are served as they are.

Capabilities

People sign in with an emailed link. Brod says the app never sees a password and keeps no user table.
Owner, editor and viewer roles are available by default. Groups defined once in the workspace are used by every app.
Each app has its own SQLite database with transactions. Each app also has key-value storage and file storage.
Secrets are set with set_secret and are write-only. Brod says nobody can read one back after it is set, the agent included.
Every deploy is kept. Rollback takes effect on the next request and rebuilds nothing.

When to use it

You built a small tool with an AI agent and want it online for yourself or your team.
You want teammates to sign in once and find a shared tool on the workspace home page.
You want your agent to search recent request logs while it fixes a bug.