InsForge is an open-source backend platform designed to be operated by AI coding agents rather than only by people. It bundles the services a full-stack app needs — Postgres, authentication, S3-compatible storage, edge functions, a model gateway, compute, and site deployment — behind interfaces an agent can drive end to end.
Agents reach it two ways. A Model Context Protocol (MCP) server, available on both self-hosted and cloud instances, exposes InsForge operations as callable tools. On the cloud, a command-line interface paired with skills lets an agent work from the terminal. Either way the agent can read backend context and state — documentation, schemas, deployed functions, bucket contents, auth configuration, runtime logs — and configure primitives directly: deploy edge functions, run migrations, create storage buckets, and set up auth providers.
You can use the hosted service or self-host with Docker Compose. A setup script writes the compose files and generates the JWT secret, encryption key, Postgres password, admin password, and access keys into a .env file; each project gets its own directory, compose project name, and ports so instances do not share containers. Storage defaults to the local filesystem, with overlays for bundled MinIO or RustFS and support for any external S3-compatible store, which also enables an S3-compatible gateway at /storage/v1/s3. One-click templates exist for Railway, Zeabur, Sealos, and RepoCloud.
Features
- Authentication: user management, sessions, and OAuth providers
- Database: a Postgres relational database with migrations an agent can run
- Storage: S3-compatible file storage, backed locally, by bundled MinIO or RustFS, or by an external S3 store
- Model gateway: an OpenAI-compatible API that routes across multiple LLM providers
- Edge functions: serverless code deployed and run at the edge
- Compute: long-running container services, in private preview
- Site deployment: build and host a site with a live URL
- MCP server: exposes backend operations as tools for any MCP-compatible agent
- CLI and skills: a cloud-only terminal path agents invoke directly
- Self-hosting: a Docker Compose stack with generated secrets and per-project isolation
- Vectors and realtime: pgvector embeddings for semantic search, plus realtime data updates
