Nylas is Good to agents.
Discry independently scored how well an AI agent can discover and understand the Nylas 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 · 79/100Comprehension
55% of score · 96/100What we found
- Nylas is among the most agent-ready APIs assessed: it ships an official MCP server (16 tools, listed on PulseMCP and Glama), maintains AGENTS.md plus a dedicated agent-skills repo (nylas/skills) that works with 40+ coding agents, and publishes both llms.txt and llms-full.txt. An agent discovering Nylas finds purpose-built entry points at every layer.
- robots.txt explicitly welcomes every major AI crawler (GPTBot, ClaudeBot, anthropic-ai, Google-Extended, PerplexityBot, CCBot, and more) and sets Content-Signal directives (search/ai-input/ai-train = yes) — an agent's crawler is never blocked or ambiguously treated.
- Comprehension is near-exemplary: endpoint guides are task-oriented, code samples use realistic values across six languages plus CLI, and 'Keep in mind' sections spell out rate limits, 503 backoff timing, idempotency, timeout requirements, and provider-specific constraints — exactly the self-correction context an agent needs to recover from failures.
- The biggest discovery gap is the absence of a true OpenAPI/Swagger spec — Nylas offers only a public Postman collection, so an agent that wants a machine-readable contract for codegen or contract testing has no OpenAPI document to pull.
- The llms.txt is well-structured and API-focused but oversized (~76KB / ~19K tokens), pushing past the right-sized threshold for a quick-orientation index — agents pay a larger token cost than necessary before reaching specific docs.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Publish a versioned OpenAPI 3.x spec at a stable URL (e.g., developer.nylas.com/openapi.json) and link it from llms.txt and the API reference. This is the single largest discovery-score lever and directly unblocks agent codegen and contract testing.
- 02Trim llms.txt to a lean index (<50KB / ~12K tokens) by moving the exhaustive link lists into llms-full.txt while keeping the 'Instructions for LLM Agents' summary and top-level section links up front, so agents orient in a few thousand tokens.
- 03Add a /.well-known/mcp.json manifest declaring the official Nylas MCP server's tools and auth, so MCP-aware agents can auto-discover the server directly instead of relying on a third-party registry lookup.
- 04Lead guide pages with an answer-first method + URL block before the 'Before you begin' prerequisites, so an agent can extract the core call without scrolling past setup context.
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 exceptionally well for a scanned API: dedicated error-code pages (200/400/500/700 responses + event codes) describe causes and recovery steps, both Nylas and provider rate limits are documented with backoff guidance, cursor-based pagination is covered in best practices, and idempotent sends via the Idempotency-Key header have their own guide. These are documentation claims only — live behavior requires a full Discry Audit.