OSA Procurement Guard
Pre-purchase trust, pricing and budget decisions for AI agents buying APIs and MCP tools.
Open source Repository Open in the app JSON README (API)
About
Pre-purchase trust, pricing and budget decisions for AI agents buying APIs and MCP tools.
Details
- Kind
- MCP servers
- Topic
- Government & public data
- Publisher
- omarfff
- Origin
- official
- Category
- ferramentas
- Transport
- http
- Version
- 0.2.2
- Stars
- 1
- Last push
- 2026-09-04T17:12:46Z
- Repository state
- ativo
- Language
- JavaScript
- Added
- 2026-08-29 04:00:58
- Updated
- 2026-08-29 04:00:58
- Origin id
io.github.omarfff/osa-procurement-guard
README
# OSA OSA is a machine-commerce trust and procurement system for evaluating agent/API endpoints before purchase or routing spend. ## Core product OSA combines live endpoint verification with historical evidence and produces a TrustScore plus confidence and reason codes. - `GET /best` — rank matching endpoints by trust, intent and optional max price. - `GET /score` — live-verify a registered endpoint. - `GET /history` — return historical snapshots. - `POST /ingest` — authenticated registry ingestion. - `POST /sources/mcp` and `POST /sources/bazaar` — ingest MCP/Bazaar-shaped records. - `/.well-known/osa.json` — machine-readable product metadata. TrustScore currently considers uptime, latency, price stability, schema stability, payment stability and transaction evidence. Confidence is reported separately. ## Architecture OSA uses a split-plane design: - **Product/Data/Commercial plane** — procurement/MCP distribution, pricing and benchmark history, endpoint observations, product usage, leads/deals and payment reconciliation. - **Control/Execution/Security plane** — VPS agents, approvals, task policy, audit/control alerts, SSH bridge and certificate-only RAW SSH execution. - **GitHub** — canonical source, CI/security scanning and controlled deployment workflows. The split keeps product data workloads separate from privileged infrastructure execution. ## Security - Endpoint verification is read-only (`GET`/`HEAD`). - SSRF defenses reject local/private/link-local/documentation targets and unsafe URL schemes. - DNS results and redirect targets are revalidated. - Response size, redirect count, timeouts and concurrency are bounded. - Ingestion is authenticated in production. - Privileged VPS execution uses short-lived SSH certificates and host-key verification. - Secrets and private keys are not committed to this repository. ## Commercial acquisition tool `npm run lead:audit -- https://example.com` The Lead-Leakage Audit performs a public-page, read-only conversion-path audit. It does not submit forms, authenticate, bypass controls or access private systems. It can be used as an evidence-first acquisition wedge for OSA services without becoming a separate product strategy. ## Quick start ```bash npm install npm test npm start ``` ## Payments Optional x402 v2 payment middleware can protect `/best` and `/score`. Payments remain disabled when `OSA_PAY_TO` is not configured. Test/sandbox payment events must never be counted as revenue. ## Operating rule OSA is one product. Agent Trust Oracle, Procurement Guard, pricing/benchmark intelligence, MCP reliability and payment-rail checks are modules of the same pre-transaction decision system. Progress means verified external usage, verified payment and repeat usage — not internal test rows, crawler probes, generated dashboards or speculative revenue. ## Revenue-first gate Before first verified external payment, OSA keeps one primary revenue path plus one backup and freezes generalized product/infrastructure/payment/browser/autonomy expansion unless it is the minimum fix for a named external blocker or a real safety incident. Every non-trivial change should carry the decision record defined in `knowledge/05-revenue-gate.md`. Pull requests are checked by the Revenue Gate workflow. The governing distinction is strict: BUILT != VERIFIED != SELLING != PAID.