API & AI

MCP Server

A headless app that exposes its tables to an LLM client over MCP. Per-entity access is declared with aiAccess; the operator activates the server with an environment flag.

aiAccess

Use this template Deploy on Scalingo View source

The configuration

This is the whole app. Copy it, change it, make it yours.

app.yaml

name: mcp-server-example
version: 1.0.0
description: A headless Sovrium app that exposes tables to an LLM client via MCP

# This is a HEADLESS app — no `pages` array. The MCP protocol is the only surface.
#
# Per-entity MCP exposure is declared via `aiAccess` on each table. The MCP server
# itself is activated by the operator via environment variables (NOT in schema):
#
#   MCP_ENABLED=true
#   MCP_TRANSPORT=streamable-http
#   MCP_AUTH_STRATEGY=oauth
#
# Schema author declares intent; operator activates. README.md ("Turning MCP on") lists the
# tools this exposes once MCP_ENABLED=true — and why there is nothing to add under config/.
#
# Sample data lives in seed/ and loads with `sovrium seed app.yaml`. There are no pages, so
# the REST API and the MCP tools are the only way to see it.

auth:
  $ref: ./config/auth.yaml

tables:
  - $ref: ./config/tables/tags.yaml
  - $ref: ./config/tables/documents.yaml

Run it

Install Sovrium, then start the app from its config.

>_ terminal

curl -fsSL https://sovrium.com/install | sh
MCP_ENABLED=true sovrium start app.yaml

Build your own

Start from a template or an empty file. One config, and Sovrium runs the whole app.

Read the docs Browse more apps