Skip localhost. Start with a real URL.
Prompt your agent and every change swaps into production in place, at a real
address. Services are Rust compiled to wasm32-wasip2, on a shared
runtime with an isolated transactional store, capability security, and
transactions that span services.
Developer hub · API reference · Agent bootstrap · boogy.ai
You don't start by learning an SDK. Install the plugin and ask your agent for what you want; it carries the skills and the rules, and it does the building.
claude plugin marketplace add Boogy-ai/boogy-superpowers
claude plugin install boogy@boogy
Then, inside Claude Code:
/reload-plugins
Do not skip that. Nothing activates until it runs, and there is no error saying so — the session simply behaves as though the plugin were never installed.
Signing in and deploying are the boogy CLI's job:
cargo install --locked --git https://github.com/Boogy-ai/boogy-sdk boogy-cli
boogy login
boogy login shows a link and a one-time code; you sign in in the browser, and
the CLI keeps the token in its own credentials file — it never passes through
your agent. A first-time account picks a handle, its username (4 to 30
characters: lowercase letters, digits and -).
That's the setup. Now ask for what you want: "build me a link shortener with a
web page, and deploy it." It goes live at an address of its own, such as
links-4k2q.boogy.app, and stays there as you iterate.
Not on Claude Code?
Vendor the skills into your project — boogy skills install, or
npx degit Boogy-ai/boogy-superpowers/skills .claude/skills — and install the
CLI as above. Or have your agent read
boogy.ai/llms.txt and follow it.
Any MCP-capable agent can also use Boogy's public, anonymous MCP server:
claude mcp add boogy https://api.boogy.ai/mcp
It serves guidance (get_started, list_skills, get_skill,
manifest_reference) and host-truth validation (validate_manifest,
check_service). It neither signs in nor deploys: that is boogy login and
boogy deploy, or the /v1 REST API.
Humans — the developer hub routes you to the docs, the SDK, and the catalog.
You write handlers. The runtime gives each deployed service:
| An address | Every service is served at the root of its own address, <name>-<suffix>.boogy.app, which stays the same as you redeploy — and a redeploy drops no requests. Bring your own domain when you want one. |
| Surfaces | REST, JSON-RPC, MCP tools, and protobuf RPC (Connect and gRPC-Web) from the same handlers. openapi.json and openrpc.json are generated and served for you. |
| Data | A private relational, ACID store. Declare models in Rust; indexes and columns reconcile on deploy. One transaction can span a whole cross-service call tree. |
| Frontend | Ship TypeScript or JS source — the platform bundles it at deploy and serves it on the same origin as your API. No Node and no build step on your machine. Static site, full-stack, or backend-only. |
| Identity | Platform accounts, "Sign in with Boogy" for your end users, a different pseudonym per service so apps can't correlate them, API keys, scopes, and on-behalf-of delegation between services. |
| Security | Capabilities are deny-by-default and declared in the manifest. Ingress mode, rate limits and CORS are host-enforced, not app code. |
| The rest | Background jobs, real-time channels over websockets, streaming responses, file storage, cryptographic signing, OAuth connections to third-party accounts, outbound HTTP behind an allowlist, and LLM completions across providers. |
Services call each other in-process via peer::fetch — no network hop, and
the caller's transaction and identity carry through.
flowchart TB
client(["Client · browser · agent"])
subgraph host["Boogy runtime"]
direction TB
edge["HTTP edge<br/>one address per service · auth · ingress · rate limit · CORS"]
sched["Fair scheduler<br/>per-owner budgets · backpressure"]
inst["Your service<br/>Rust to wasm32-wasip2"]
fe["Static assets<br/>served by the host"]
edge --> sched --> inst
edge --> fe
end
store[("Per-service<br/>transactional store")]
peer["Other services<br/>(the mesh)"]
ext["Third-party APIs"]
client -->|"request"| edge
inst -->|"REST · JSON-RPC · MCP · protobuf"| client
inst -->|"websockets · SSE"| client
inst -->|"store"| store
inst -->|"peer::fetch"| peer
inst -->|"outbound_http · OAuth connections"| ext
| Repo | What it is |
|---|---|
| boogy-sdk | The SDK + boogy CLI — Router, typed handlers, wit_glue!, store helpers, McpServer, the WIT interfaces. Full architecture overview and API reference. |
| boogy-web | The browser SDK (@boogy/web) — add user sign-in ("Sign in with Boogy") and authenticated requests to your app's frontend with one fetch(). |
| boogy-superpowers | The Boogy skills, installed as the boogy plugin: 51 skills that teach a coding agent to build Boogy services well — design, data modeling, auth, jobs, frontends, custom domains, and the platform's limits. |
| boogy-catalog | Provisionable, best-practice services — email, payments, custodial wallets, governance — you deploy into your own tenant with your own keys. |
MIT OR Apache-2.0