talonicdev is Agent-Ready to agents.
Discry independently scored how well an AI agent can discover and understand the talonicdev 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 · 95/100Comprehension
55% of score · 96/100What we found
- The scanned domain talonic.dev does NOT resolve (NXDOMAIN) — an agent handed that URL hits a dead end. The live product resolves at talonic.com (platform + docs) and api.talonic.ai (interactive Swagger). This scan assessed the real, working Talonic documentation.
- Talonic is close to a reference implementation of agent-native discovery: a root openapi.json (OpenAPI 3.1), an llms.txt with a dedicated 'For agents' section, and a comprehensive llms-full.txt spanning company, four-phase pipeline, full API contract, auth scopes, and pricing. An agent can bootstrap the entire API from a single fetch.
- First-party, hosted MCP server (mcp.talonic.com/mcp, zero-install) exposing nine tools and two resources, publicly listed on both PulseMCP and Glama and shipped as @talonic/mcp with an AGENTS.md in the repo — an agent can integrate via MCP without ever touching REST.
- Error-recovery guidance is best-in-class for agent self-correction: per-error recovery steps ('Supply one of: file, file_url, or document_id'), an explicit retry strategy with exponential backoff and jitter, Retry-After semantics, and a clear 'do not retry 400/401/403/422' rule.
- robots.txt is explicit opt-in for essentially every major AI crawler (GPTBot, ClaudeBot, anthropic-ai, Google-Extended, CCBot, PerplexityBot, Amazonbot and more), so retrieval and training crawlers are welcomed rather than merely tolerated.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Redirect or claim talonic.dev — the domain currently returns NXDOMAIN, so any directory entry, backlink, or agent using the .dev URL fails before discovery begins. A 301 to talonic.com would recover that traffic instantly.
- 02Fix /.well-known/mcp.json — it currently returns HTTP 500. Publishing a valid MCP manifest there would let agents auto-discover the already-excellent hosted MCP server via the standard well-known path, closing the only failed discovery signal.
- 03Reconcile the error contract across sources: the docs page shows a 3-field envelope ({status, code, message}) while llms-full.txt documents an 8-field camelCase-leaning envelope and contradicts it on the 429 Retry-After header. Publish one authoritative shape so agents can parse errors deterministically.
- 04Publish the Python SDK (currently 'in development') or add multi-language client snippets on individual endpoint pages — the OpenAPI spec enables codegen, but first-class examples reduce agent trial-and-error for the largest non-JS user base.
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.
Execution characteristics are documented to an unusually high standard: machine-readable JSON error envelope with a full code table, per-tier and per-key rate limits with X-RateLimit-* headers, cursor pagination on all list endpoints, and Idempotency-Key support (24h TTL) with a dedicated page. One caveat an agent could hit: the error envelope shape differs between the docs page ({status, code, message}) and llms-full.txt (8-field {statusCode, code, error, message, retryable, request_id, timestamp, path}), and the two sources disagree on whether 429 responses carry a Retry-After header.