Lula is Good to agents.

What we found

• —lula.com no longer serves the 'Lula' insurance API — it 301-redirects to Gail (meetgail.com, Nothing Technologies Inc.), a conversational-AI platform for insurance, banking, and lending. An agent following the registry's 'Lula' entry lands on a different company than the name implies; this assessment scores Gail's public External API, which is what actually lives at the domain.
• —Strongly positive: an agent that reaches docs.meetgail.com finds near-ideal machine consumption — a published OpenAPI 3.1 spec (JSON + YAML), an API-focused llms.txt with an explicit 'Instructions for AI Agents' section, clean per-page Markdown (append .md to any URL), and a hosted MCP server at /_mcp/server. These are Fern-generated docs built for agents, not just humans.
• —Endpoint descriptions are task-oriented ('Start a single outbound call', 'Download a CSV report of campaign results, including outcomes and transcripts') and every endpoint ships copy-pasteable SDK snippets in 8 languages, so an agent can learn both WHAT an endpoint does and HOW to call it without guessing.
• —Weak spot for execution: examples use placeholder values ("to": "string", empty {} request bodies) rather than realistic phone numbers/IDs, and there is no documented rate-limiting or per-error recovery guidance — an agent would learn valid inputs, request limits, and retry behavior only by failing against the live API.
• —Discovery is split-brain: the docs subdomain welcomes AI crawlers, but the primary brand domain (lula.com -> meetgail.com) robots.txt hard-blocks GPTBot, anthropic-ai, Google-Extended, CCBot, Bytespider and cohere-ai. An agent starting from the brand domain is turned away before it ever finds the excellent docs subdomain.

What to change

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

1. 01Replace placeholder example values ('string', empty {}) with realistic ones — E.164 phone numbers, sample UUIDs, plausible campaign/script names — so agents can adapt the copy-paste examples without having to invent valid inputs.
2. 02Document rate limits (numeric limits plus the response header names an agent should read) and add per-error recovery guidance for the top 400/401/429 cases, turning the already-structured ProblemDetails schema into actionable self-correction.
3. 03Align the marketing-domain robots.txt with the docs: stop blocking GPTBot/anthropic-ai/CCBot etc. on meetgail.com (or at minimum allow them to reach the docs) so agents can discover the API starting from the brand domain.
4. 04Add a /.well-known/mcp.json manifest advertising the existing /_mcp/server endpoint, and list the MCP server in public registries (Glama, Smithery, PulseMCP) so agents can find it without first reading llms.txt.
5. 05Fix the directory identity: correct the registry entry (and add an AGENTS.md) so the API is discoverable under its real name, Gail / meetgail.com, rather than the stale 'Lula' brand at a redirecting domain.