# Serge

> Agent readiness platform. Scans APIs and scores how ready they are for AI agents.

## What Serge does

Serge scans any domain across 6 layers — Discovery, Schema, Protocol, Auth, Behavior, and Pricing — and returns an agent readiness score out of 100. Each of the 35 checks passes, fails, or warns, and every failure includes a specific fix-it recommendation.

The Agent Readiness Stack is the framework that defines what "agent ready" means. Serge created this standard and provides the benchmark data.

## Products

- [Serge Scanner](https://serge.ai): Free agent readiness score for any domain
- [Serge API](https://serge.ai/docs): REST API to initiate scans and retrieve results programmatically
- [Serge MCP](https://serge.ai/docs/mcp-server): Claude Desktop extension — AI agent benchmarking for websites. Instruments every browsing step with timing, screenshots, and accessibility analysis.

## API

- Endpoint: POST https://serge.ai/api/scan — initiate a domain scan
- Endpoint: GET https://serge.ai/api/scan/{id} — retrieve scan results
- Authentication: Bearer token via `Authorization: Bearer sk_serge_...` header
- Anonymous: 10 scans/hr per IP. Authenticated: 60 scans/hr per workspace.
- Create secret keys at https://serge.ai/member/api-keys (free account required)
- [API Documentation](https://serge.ai/docs)
- [OpenAPI Spec](https://serge.ai/openapi.json)

## The 6 Layers

1. Discovery — Can agents find you? (llms.txt, agent cards, JSON-LD, sitemap)
2. Schema — Can agents understand your API? (OpenAPI spec, error schemas, rate limits)
3. Protocol — Can agents interact? (MCP server, A2A card, SDKs, webhooks)
4. Auth — Can agents authenticate? (OAuth M2M, API keys, scoped permissions)
5. Behavior — Is the API agent-friendly? (idempotency, structured errors, retry-after)
6. Pricing — Is there agent-friendly pricing? (machine-readable pricing, free tier)

## Serge MCP (Claude Desktop extension)

Serge MCP is an MCP server for Claude Desktop that benchmarks websites for AI agent accessibility. It instruments every browsing step — clicks, searches, form fills — with timing, screenshots, and accessibility analysis. 8 tools: serge_start_session, serge_navigate, serge_read_page, serge_click, serge_type, serge_scroll, serge_screenshot, serge_end_session. No API keys needed — works on your existing Claude subscription.

- Documentation: https://serge.ai/docs/mcp-server
- Install: Add `@serge-ai/mcp-server` in Claude Desktop config
- GitHub: https://github.com/SuperstellarLLC/serge-mcp-server

## Extended documentation

For comprehensive documentation including all 35 checks, scoring methodology, and API usage examples, see [llms-full.txt](https://serge.ai/llms-full.txt).

## SergeBot

SergeBot is the crawler that powers the scanner. It identifies itself as `SergeBot/1.0` and only requests well-known agent-facing files. It does not scrape, index, or store page content.

- Bot documentation: https://serge.ai/bot
- Contact: bot@serge.ai

## Company

Serge is built by Superstellar LLC, based in Zug, Switzerland.

- Website: https://serge.ai
- Contact: api@serge.ai