# SAM Suite — agent integration

> SAM is a guardrail layer for autonomous agents: a hosted REST suite your agent calls over plain HTTP/JSON with `Authorization: Bearer <key>`. No SDK — any runtime that can send an HTTP request works (OpenAI, Cursor/MCP, custom). SAM is non-custodial: it never approves or holds your money, it only enforces the spend dials you set; funds move on your own rails.

Auth: `Authorization: Bearer <api_key>`, scoped per tenant + permission server-side (a smuggled `tenant_id` in a body is ignored). `GET /healthz` → liveness + honest leg accounting (`legs_mounted` / `legs_missing`).

## Start here
- [Agent Integration Contract (full)](/llms-full.txt): the one page to integrate — leg catalog, required perms, enums, envelopes, idempotency, mandate matching.
- [Connect an agent (web page)](/developers): the same surface, human-readable.
- [Health](/healthz): liveness + legs_mounted / legs_missing.

## Legs (HTTP path prefixes)
- SAMShield `/shield` — scan untrusted content for prompt-injection / policy risk
- SAMHandler `/handler` — provision agent identities (+ seat payment)
- SAMScope `/scope` — issue short-lived scoped credentials
- SAMcypher `/cypher` — governed payments / transfers / subscriptions
- sam-license `/license` — validate a SAMSerum / SAMShredder license (gate-on-use)

## Examples (runnable, no SDK)
- [reference_client.py](/agent-examples/reference_client.py) — stdlib urllib only — the no-SDK proof
- [raw_http_curl.sh](/agent-examples/raw_http_curl.sh) — curl — any shell-capable agent
- [openai_tool_calling.py](/agent-examples/openai_tool_calling.py) — OpenAI function/tool-calling (schema + dispatch)
- [mcp_tool.py](/agent-examples/mcp_tool.py) — MCP server (Cursor / Claude Desktop) — the MCP→HTTP bridge