What is InsForge?
InsForge is a backend platform in the mould of Supabase, designed so that a coding agent rather than a person does most of the setup. A project bundles a Postgres database with pgvector and row-level security, user authentication with OAuth and magic links, S3-compatible file storage, realtime channels, Deno edge functions, an OpenAI-compatible model gateway, frontend deploys and, in private preview, long-running containers.
Agents work it through an MCP server, which exposes tools such as get-table-schema, run-raw-sql, create-function, create-bucket, create-deployment and get-container-logs, or through a CLI paired with agent skills, which is cloud only. The project is Apache 2.0, about 13,200 GitHub stars, and shipped v2.3.5 on 23 October 2026.
It can be self-hosted with Docker Compose or run on InsForge Cloud. Two parts lean on other companies: the model gateway routes through OpenRouter, and Sites deploys frontends to Vercel.
Is InsForge worth it?
InsForge's useful idea is that a coding agent should be able to read the backend's state, not just write code against it. Its MCP server returns schemas, deployed functions, bucket contents and container logs, so when a generated app fails the agent can look at what actually happened rather than guess. Supabase's MCP server does much of the same for Supabase, so the real question is whether you want a younger, Apache-licensed platform shaped around that workflow.
The cloud pricing follows the familiar backend-as-a-service pattern of a generous free tier and metered overage. Free gives 50,000 monthly active users, a 500 MB database, 1 GB of storage and 100,000 function calls, but pauses projects after a week of inactivity. Pro is $25 a month, or $20 billed yearly, with 100,000 users, an 8 GB database, 250 GB of bandwidth and 100 GB of storage, then $0.125 per extra database GB and $0.09 per bandwidth GB. Each project's server is billed separately from $5 a month for the Nano size, with a $10 monthly credit per organisation. SOC 2, SSO and a paid HIPAA add-on are Enterprise only.
Self-hosting is straightforward but the defaults need attention. The setup script generates the JWT secret, encryption key and database password into a mode-600 .env file; building from source instead falls back to published default secrets for any you leave unset, which is dangerous on anything internet-facing. Self-hosted instances send anonymous telemetry, a start event and a daily heartbeat with version and which features were touched, until you set INSFORGE_TELEMETRY_DISABLED=1. The run-raw-sql tool needs the admin key and blocks system tables, but an agent holding that key can still drop your own tables.
The project is 15 months old with 236 open issues and pull requests and four releases since September, so expect movement. It is a reasonable backend for prototypes and small apps built mostly by agents; a team with production data and an existing Postgres habit gains less from it than from giving the agent a well-scoped database server.
Where it earns its place
A solo founder building an MVP with Claude Code or Cursor
Get a database, logins and file uploads without setting them up by hand
Create a cloud project, paste the MCP config into the editor and let the agent create tables, buckets and functions. Free covers a prototype; Pro at $25 a month avoids the one-week inactivity pause.
A two-person agency shipping small client apps
Give each client a separate backend we can hand over
Run one self-hosted instance per client directory with its own COMPOSE_PROJECT_NAME and ports, or a cloud project each; two Nano projects on Pro come to $25 a month once the $10 compute credit is applied.
A platform engineer asked to approve agent access to a backend
Limit what a coding agent can do to production data
Keep the admin key out of the agent's MCP config for production, since run-raw-sql and bulk-upsert need it, and use backups and forward-only migrations so a bad change can be rolled back.
Before you commit
- Setup
- Self-host: run the setup script, edit API_BASE_URL in .env and docker compose up -d, then open localhost:7130 to connect the MCP server. Cloud: npx @insforge/cli login and link a project, about five minutes per the docs.
- Integrations
- MCP server for Cursor, Claude Code, Windsurf, Codex, Copilot and others; TypeScript and Swift SDKs; REST API; Stripe and Razorpay payments; OpenAI-compatible gateway routed through OpenRouter; frontend deploys through Vercel.
- Data and privacy
- Self-hosted data stays on your servers apart from anonymous telemetry, which INSFORGE_TELEMETRY_DISABLED=1 stops; the cloud sends none. Model calls go to OpenRouter and site builds to Vercel. Apache 2.0; no GitHub security advisories published.
- Cost at scale
- Pro is $25 a month plus overage: 20 GB of database and 500 GB of bandwidth add about $1.50 and $22.50. Compute ranges from Nano at $5 a month to 16XLarge at $2,560 a month per project.
- Limits
- Free projects pause after a week idle; Compute is private preview; Sites functions run in Vercel's iad1 region by default; 236 open issues and pull requests.
What's good
- MCP tools let agents read schemas, logs and deployed state
- Postgres with pgvector, auth, storage and functions in one project
- Apache 2.0 and self-hostable with Docker Compose
- Cloud Pro at $25/month with clear overage rates
What's not
- Building from source falls back to default secrets if unset
- Model gateway needs OpenRouter; Sites deploys to Vercel
- Self-hosted telemetry on by default
- SOC 2 and SSO only on Enterprise; CLI skills are cloud only
Best for
Solo developers and small teams who let a coding agent build full-stack prototypes and want the agent to manage the backend too.
InsForge: common questions
- Is InsForge worth it?
- A sensible bet that agents need a backend they can inspect and change themselves, held back for now by its youth and by leaning on Vercel and OpenRouter for two of its parts. We rate it 3.8 out of 5.
- How much does InsForge cost?
- Apache 2.0 and free to self-host. InsForge Cloud: Free; Pro $25/month or $20/month billed yearly, plus compute from $5/month per project with a $10 monthly credit; Enterprise custom.
- What are the downsides of InsForge?
- Building from source falls back to default secrets if unset; model gateway needs OpenRouter; Sites deploys to Vercel; self-hosted telemetry on by default; sOC 2 and SSO only on Enterprise; CLI skills are cloud only.
- Who is InsForge best for?
- Solo developers and small teams who let a coding agent build full-stack prototypes and want the agent to manage the backend too.
- How much does InsForge cost for a team or in production?
- Pro is $25 a month plus overage: 20 GB of database and 500 GB of bandwidth add about $1.50 and $22.50. Compute ranges from Nano at $5 a month to 16XLarge at $2,560 a month per project.
- Is InsForge safe for company data?
- Self-hosted data stays on your servers apart from anonymous telemetry, which INSFORGE_TELEMETRY_DISABLED=1 stops; the cloud sends none. Model calls go to OpenRouter and site builds to Vercel. Apache 2.0; no GitHub security advisories published.
Answers checked Oct 26, 2026.