Build on VibeInfra
VibeInfra runs production-realistic SRE incident simulations in live disposable sandboxes. Everything the platform knows about its incident catalog is available over a public REST API, an OpenAPI description, an MCP server for AI agents, and SDKs for JavaScript and Python.
The catalog surface needs no API key. It is public by design — it is the platform's index — so an agent or a script can read it with a single unauthenticated request.
Thirty-second start
# Every incident in the catalog, as JSON
curl -s https://api.vibeinfra.id/api/v1/courses
# One incident's full blueprint
curl -s https://api.vibeinfra.id/api/v1/courses/incident-22-k8s-crashloop
# Is the sandbox pool busy right now?
curl -s https://api.vibeinfra.id/api/v1/labs/capacityDeveloper resources
API reference
Every endpoint, its parameters and its response shape — catalog, glossary, capacity, telemetry, leaderboard, credit packs and learner progress.
Read the API docs →Machine-readable spec
The complete API description, ready to paste into a generator, an agent, or Postman. Served as JSON and YAML.
openapi.json →openapi.yaml →
MCP server
Give an AI agent five read-only tools over the catalog, live capacity and the glossary. Runs over stdio, no credentials.
npx -y @vibeinfra/mcp-serverJavaScript SDK
Typed client with zero runtime dependencies. Works in Node 18+ and any runtime with a global fetch.
npm install @vibeinfra/sdkPython SDK
Standard-library-only client, so it adds no transitive dependencies to an agent runtime or a CI image.
pip install vibeinfraSRE incident drill skill
A packaged skill that teaches an agent to pick the right incident for a symptom and skill tier, then coach the diagnosis without giving away the fault.
Read SKILL.md →Machine-readable discovery
Every resource above is also advertised in standard discovery manifests, so an agent can find them without reading this page.
| Path | Standard | What it lists |
|---|---|---|
/.well-known/ard.json |
Agentic Resource Discovery 1.0 | Every agent-facing resource VibeInfra publishes: MCP server, OpenAPI description, agent skill, catalog feed, llms.txt and both SDKs. |
/.well-known/ai-catalog.json |
ARD (pre-rename path) | Byte-identical to ard.json, for consumers that only know the earlier path. |
/.well-known/mcp.json |
MCP server.json |
The server card for @vibeinfra/mcp-server: package, transport and configuration. |
/openapi.json |
OpenAPI 3.1 | The full REST contract. |
/llms.txt |
llms.txt | Plain-language site summary and a link to every incident. Long form at /llms-full.txt. |
/sitemap.xml |
Sitemap | Every indexable page, including this one. |
Fair use
Read endpoints are unmetered within reason. The write paths carry per-caller limits, because
each sandbox start provisions real containers: 5 lab starts and 12 grading checks per minute,
5 waitlist submissions per minute per IP. A throttled request comes back as
429 — back off rather than retrying immediately.
Crawling is welcome; the catalog is meant to be indexed. See robots.txt for the crawler policy.
Questions
Open an issue on GitHub, or email [email protected].