Expose your tables to an LLM over MCP.
Built with Sovrium — a configuration-as-code interpreter: one config file in, a complete self-hosted web application out. This is a headless template — its surface is the REST API and the MCP endpoint, not web pages.
Click Use this template on GitHub to copy this app into your own repository (clean history, yours to modify), or scaffold it locally:
curl -fsSL https://sovrium.com/install | sh
sovrium init my-mcp-server --template mcp-serverTables with per-entity aiAccess exposure, served to LLM clients over the Model Context Protocol when MCP_ENABLED=true.
Everything is declared in app.yaml and the config/ tree —
no application code. Edit the config, restart, done.
sovrium start app.yamlZero-config: embedded SQLite, local file storage, no env vars required to boot. See
.env.example for the optional variables (database, auth bootstrap,
email, AI).
The Deploy on Scalingo button above provisions the app with a PostgreSQL addon
(Scalingo's filesystem is ephemeral — the database keeps your data across deploys; file
uploads are stored in Postgres too). Secrets are generated automatically; you only fill in
BASE_URL. Any other host works the same way: run the sovrium binary with this config
(see DEPLOY.md).
This repository is auto-published from the
Sovrium monorepo on every release
(source: templates/mcp-server).
Issues are welcome here; please send code contributions upstream to the monorepo so the
template stays in sync with the engine. The pinned engine release lives in
.sovrium-version.
License: MIT. The Sovrium engine itself is licensed separately (BSL 1.1).