# Knostic

> Knostic secures the AI agent supply chain. It discovers shadow AI, coding assistants,
> MCP servers and IDE extensions inside an enterprise, and scans them for prompt
> injection, data exfiltration, secret leakage and destructive commands. Its public
> reputation service, AgentMesh, exposes a catalog of ~80,500 agent skills, ~4,800 MCP
> servers and ~59,900 IDE extensions with per-version security scan verdicts, over a REST
> API whose read endpoints serve anonymous callers.

Generated 2026-07-19 by the API Evangelist enrichment pipeline. Knostic does not publish
its own llms.txt (probed: 404 on www.knostic.ai, SPA catch-all on agentmesh.knostic.ai).

## APIs

- [AgentMesh API](https://agentmesh.knostic.ai/api): Reputation and threat intelligence for
  agent skills, MCP servers and IDE extensions. Base URL `https://agentmesh.knostic.ai/api`.
  12 operations. Bearer API key; catalog reads also answer anonymously (`X-Tier: anonymous`).

### Operations

- `GET /skills` — search the agent-skill catalog (q, status, source, hash, sort, page, limit)
- `GET /skills/{skill_id}` — skill detail with version history and scan results
- `GET /skills/{skill_id}/versions/{version_id}` — scan results for one skill version
- `GET /mcp` — search the MCP server catalog
- `GET /mcp/{server_id}` — MCP server detail with versions and scan results
- `GET /mcp/{server_id}/versions/{version_id}` — scan results for one server version
- `GET /extensions` — search IDE extensions with risk assessment (q, marketplace, status, hash)
- `GET /extensions/{ext_id}` — extension detail with risk level, reasons and code evidence
- `POST /scan/skill/{slug}` — trigger an on-demand skill scan (API key; 30/day)
- `POST /scan/extension/{slug}` — trigger an on-demand extension scan (API key; 30/day)
- `GET /scans` — your scan history; carries `X-Scan-Limit-Remaining`
- `GET /scans/{scan_id}` — poll a scan to completion

## Specs

- [OpenAPI 3.1](openapi/knostic-agentmesh-openapi.yml): reconstructed from Knostic's own
  published endpoint catalog; Knostic publishes no OpenAPI
- [Overlay](overlays/knostic-agentmesh-overlay.yaml): API Evangelist enhancements
- [Data model](data-model/knostic-data-model.yml): entity graph
- [Authentication](authentication/knostic-authentication.yml)
- [Conventions](conventions/knostic-conventions.yml)
- [Error catalog](errors/knostic-problem-types.yml)
- [Rate limits](rate-limits/knostic-rate-limits.yml)
- [Conformance](conformance/knostic-conformance.yml)
- [Lifecycle](lifecycle/knostic-lifecycle.yml)
- [MCP candidate surface](mcp/knostic-mcp.yml)
- [Agent Skills](skills/_index.yml)

## Key semantics

- Errors are a flat `{"detail": "..."}` object, not RFC 9457 problem+json.
- Pagination is page-number: `page` + `limit` (max 200), with `total`/`page`/`limit` returned.
  The result array is keyed per resource: `skills`, `mcpServers`, `extensions`, `scans`.
- Scans are asynchronous submit-and-poll. There is **no idempotency key** — re-issuing a
  scan POST creates a new job and spends another of the 30 daily credits.
- `unscanned` does not mean safe; it means nobody has looked.
- The `status` filter vocabulary (dangerous/risky/safe/unscanned) differs from the values
  returned on rows (pass/warn/fail/gray, safe/low/medium/high/critical).
- `fileSha256` is the cross-collection join key — every collection accepts a `hash` filter,
  so artifacts are addressable by content rather than by name.

## Open source

- [AgentSonar](https://github.com/knostic/AgentSonar): Go CLI that detects shadow AI by
  monitoring outbound traffic and scoring process-to-domain pairs
- [OpenAnt](https://openant.knostic.ai): LLM-based vulnerability discovery — stage 1
  detects, stage 2 attacks, what survives is a verified finding
- [MCP-Scanner](https://github.com/knostic/MCP-Scanner): Shodan-based scanner for exposed
  MCP servers
- [openclaw-shield](https://github.com/knostic/openclaw-shield): five-layer security plugin
  for OpenClaw agents
- [openclaw-telemetry](https://github.com/knostic/openclaw-telemetry): agent observability
  with redaction, hash-chain integrity and syslog/CEF SIEM forwarding
- [openclaw-detect](https://github.com/knostic/openclaw-detect): MDM detection scripts
- [open-tools](https://github.com/knostic/open-tools): vulnerable-npm-package scanner and
  GlassWorm YARA suite
- [extension-check-skill](https://github.com/knostic/extension-check-skill): Knostic's
  published Agent Skill for auditing installed editor extensions

## Docs

- [Website](https://www.knostic.ai/)
- [AgentMesh](https://agentmesh.knostic.ai/)
- [API reference](https://agentmesh.knostic.ai/api)
- [Blog](https://www.knostic.ai/blog)
- [Glossary](https://www.knostic.ai/glossary)
- [Security program](https://www.knostic.ai/security)
- [Trust center](https://security.knostic.ai/) — SOC 2 Type 2
- [Terms of use](https://www.knostic.ai/terms-of-service)
- [Privacy policy](https://www.knostic.ai/privacy-policy)
- [Kirin](https://www.getkirin.com/) — AI coding safety, free up to five users
- [GitHub](https://github.com/knostic)

## Not published

Recorded as verified absences, not gaps to fill: no OpenAPI, no llms.txt, no
`/.well-known/` surface (no security.txt, OIDC or api-catalog), no status page, no
deprecation policy, no SLA, no API changelog, no webhooks or AsyncAPI event surface, no
OAuth scopes, no client SDKs on any package registry, no vulnerability disclosure program,
and no public pricing page.