Kloudle is Good to agents.

What we found

• —An agent looking for Kloudle finds it easily and can start using it with zero setup: a well-structured llms.txt, a valid /.well-known/mcp.json declaring the search/get tools, an AI-bot-friendly robots.txt, and live listings on Smithery (83/100), the MCP Registry, and claudemarketplaces all point to the same streamable-HTTP MCP endpoint with 50 free calls/day and no API key.
• —Kloudle is architected agent-first rather than retrofitted: the /agents page documents the exact search()/get() call pattern, ships a copy-paste MCP config, and returns SHA256-verified, cosign-signed static binaries with structured exit codes so an agent can branch logic without parsing stdout. This is why comprehension scores an A (90).
• —The scan penalizes Kloudle on discovery signals that assume a traditional REST API: there is no OpenAPI/Swagger spec (the highest-weighted discovery check) because the interface is MCP tools plus binaries, and no llms-full.txt. This drops discovery to a D (57) despite the surface itself being highly agent-usable.
• —Capability boundaries are exceptionally clear for an early-stage product: the docs explicitly distinguish the 7 agent scanners / 24 checks from the full platform's 681 AWS checks, label every platform surface as Live / In development / Roadmap, and state rate limits and read-only IAM scoping up front — an agent will rarely discover a limit by failing.
• —No AGENTS.md exists in the github.com/Kloudle organization, a missed quick win for a company whose entire positioning is building software with coding agents.

What to change

Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.

1. 01Publish an OpenAPI 3.x spec for the MCP tool surface (or at minimum a machine-readable tool schema linked from mcp.json) — even a two-operation (search, get) spec would flip the single highest-weighted discovery check from fail to pass and is the largest single score lever available.
2. 02Add an AGENTS.md to the primary public repo in github.com/Kloudle (e.g. the scanner-binary or MCP-server repo) covering how a coding agent should discover, download, verify, and run the scanners — a near-zero-effort fix that matches Kloudle's own AI-software-factory positioning.
3. 03Generate an llms-full.txt that expands the existing llms.txt with the full search/get call semantics, the 24 checks per scanner, exit-code contract, and the waitlist POST schema, so an agent can absorb the complete surface in one fetch.
4. 04Add explicit error-recovery guidance to the /agents docs beyond exit-code meanings: what an agent should DO on exit code 2 (re-check IAM credentials / required read-only permissions) and exit code 3 (retry with backoff), turning error documentation from descriptive into actionable.
5. 05Document a second end-to-end workflow beyond the happy-path Discover-Get-Run (e.g. multi-scanner sweep, or promoting a finding to the ledger) so multi-step workflow coverage moves from a single flow to the 2-3 chained guides agents rely on.