Labsco
MCP SERVER

Butterbase

by butterbase-ai ↗

A self-hosted backend — Postgres with row-level security, auth, storage, functions, realtime, RAG and an LLM gateway — with every one of those surfaces exposed to agents as MCP tools.

PaaS Deployment & App Hosting
Summary
An agent can build and operate a whole backend, not just query one.

Schema, row-level security, functions, storage, realtime and RAG all sit behind the same MCP surface, so the model can provision the thing it needs instead of asking you to open a dashboard first.

What it is

The runtime data plane of Butterbase, an open-source backend-as-a-service. A Fastify control API fronts three Postgres planes, a Deno runtime for serverless functions, S3/R2 storage and a Python agent runtime, and embeds an MCP server at `/mcp`.

What you get
  • Data: per-app Postgres databases with declarative schema and automatic REST endpoints, first-class row-level security policy management, a quota-protected key-value store with TTL and audit trail, and S3/R2-backed file storage with presigned URLs
  • Compute: TypeScript serverless functions on the Deno runtime, durable per-key actors for chat rooms and rate limiters, WebSocket subscriptions to table changes, edge SSR for Next.js / Remix / Astro, and static or build-from-source frontend hosting with custom domains
  • AI: a single gateway endpoint for chat, embeddings and model listing, plus managed RAG collections with document ingestion, semantic search and synthesized answers
  • Identity and ops: email and OAuth sign-in (Google, GitHub, Apple, X), JWT tuning, service keys, structured audit logs and outbound webhooks
  • All of it reachable by an agent — the same capabilities are exposed as MCP tools over HTTP at `/mcp` or over stdio via `npx @butterbase/mcp`
Requirements

Docker, Node 22+ and npm. Clone with `--recurse-submodules`, because a plain clone leaves `packages/plugin/` empty and `npm install` silently skips that workspace. Then `npm ci`, `cp .env.example .env`, and `docker compose -f docker-compose.local.yml up -d`; the first run builds images and can take several minutes. Schema is not applied on container start — run `npm run migrate:all`, then `npm run seed:dev` to create the local dev user. The control API answers on http://localhost:4000 and MCP on http://localhost:4000/mcp. Self-hosting runs the AI gateway without upstream router adapters and with a no-op billing provider; those live in the managed offering, and you wire your own through the `BillingProvider`, `QuotaEnforcer` and `RouterAdapter` interfaces.

Setup effort

One command — npx @butterbase/mcp