Brod Lets AI Agents Deploy Team Tools Through MCP

Summary

Brod is an AGPL-3.0-only, self-hosted platform that gives AI agents a place to deploy small internal team applications over MCP. A team can run it with Docker Compose, create an administrator sign-in link, generate a one-time token, connect clients such as Claude Code, Claude Desktop, or Cursor, and ask an agent to deploy an application. Brod is intended for tools such as an on-call rota, stock checker, or spreadsheet replacement, with a shared workspace identity instead of separate login and user-table implementations for every app. Workspaces contain people, groups, apps, and roles; each app can be private, shared with selected people or groups, open to the workspace, password-protected, or public. Its MCP interface exposes 13 tools covering deployment, file access, logs, secrets, access control, invitations, and domains, while widening access or inviting people requires human confirmation. Brod supports static apps and server apps whose JavaScript fetch handlers have SQLite, key-value, and file storage. Changes go to a staging address before publishing, and publishing or rollback moves a pointer that takes effect on the next request. The platform also provides conflict protection for concurrent agent edits, custom domains with HTTPS, per-app logs and analytics without cookies, and encrypted-at-rest secrets available only to server code. Its Docker-based architecture combines a control plane, Postgres, file storage, and a workerd runtime on the operator’s server. The project distinguishes this self-hosted edition from brod.sh, which runs the same image as a hosted service with plans and limits. The repository is free to run and modify under AGPL-3.0-only, and its documentation covers production operations, security boundaries, and roadmap items.