apacta.com is Poor to agents.
Discry independently scored how well an AI agent can discover and understand the apacta.com API from what’s public — not whether it’s usable. Below: every signal we checked, what’s costing the score, and what to change.
SCORED UNDER RUBRIC 1.2 · A full re-launch under Discry Score 2.5 — a new behavioral instrument, not comparable to these scores — is in progress.
Discovery
45% of score · 43/100Comprehension
55% of score · 54/100What we found
- An agent can fully machine-read this API: a complete, versioned OpenAPI 3.0 spec (v0.0.43, Partner API) is publicly downloadable as JSON and YAML at apidoc.apacta.com, mirrored on SwaggerHub, apis.guru, and an autogenerated TypeScript SDK (@apacta/sdk) on npm — the strongest agent-readiness signal Apacta offers.
- An agent discovering Apacta by URL gets no agent-native guideposts: llms.txt, llms-full.txt, and .well-known/mcp.json all 404, there is no MCP-registry listing, and the GitHub org holds only archived CakePHP/Dropbox forks with no AGENTS.md.
- The Scalar-rendered reference is answer-first and consistent — each endpoint leads with method, URL, auth requirement, params, and a response schema, and the whole surface uses uniform snake_case fields, ISO-8601 datetimes, and UUIDs — so an agent parses individual calls cleanly.
- The docs are reference-only: there are no end-to-end workflow guides showing how to chain operations (e.g. create contact → create project → log time entry → raise invoice), so an agent must infer multi-step tasks from isolated endpoints.
- Error docs list codes (401/403/404/422) with a well-structured validation payload but give no recovery guidance, and there are no documented rate limits — an agent would discover throttling and most failure fixes only by hitting them.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Publish an llms.txt at apacta.com (and/or apidoc.apacta.com) that titles the Partner API, summarizes its core resources, and links to the OpenAPI spec and key endpoints — the single fastest discovery win given the spec already exists.
- 02Add 2-3 multi-step workflow guides to the reference (e.g. 'Register work: create contact → create project → post time entries → generate invoice') so agents can chain operations instead of reverse-engineering them from endpoint stubs.
- 03Document rate limits and rate-limit response headers, plus per-error recovery steps (what a 401 vs 403 means, how to fix a 422 by field), so agents can self-correct rather than fail blindly.
- 04Enrich endpoint descriptions with task context (what each call accomplishes and when to use it) and replace empty-string request-body examples with realistic sample values to move endpointDescriptions and realisticExamples from partial to pass.
- 05List Apacta in a major MCP registry (Glama/Smithery/PulseMCP) or ship an MCP server — the org already maintains an 'apacta-claude-marketplace' repo, so surfacing an agent-facing entry point would capture the mcpRegistry and wellKnownMcp signals.
Execution coverage · INFORMATIONAL, UNSCORED
Whether an agent can actually complete a call and recover from errors is the deeper Audit layer — documented here, but not part of the Discry Score.
Machine-parseable JSON error envelope ({success, data:{code, url, message, errorCount, errors}}) is documented with a 422 validation example that names the offending field. Auth is a bearer/API-key token. Pagination is page-number based (?page={n}) with a pagination object in responses. No rate limits, rate-limit headers, or idempotency keys are documented.