Weaviate is Good to agents.
Discry independently scored how well an AI agent can discover and understand the Weaviate 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 · 71/100Comprehension
55% of score · 96/100What we found
- An agent evaluating Weaviate lands well: robots.txt explicitly allow-lists GPTBot, ClaudeBot, anthropic-ai, PerplexityBot and Applebot-Extended, and a 36KB llms.txt delivers a TL;DR, use-case fit, and an explicit 'when to use alternatives' section — so a model can grasp what Weaviate is and when to recommend it in well under 5K tokens.
- Comprehension is near-exemplary: every doc page carries an 'LLM/AI Agent Notice' pointing to llms.txt plus a 'Copy page' markdown affordance, endpoint descriptions are task-oriented ('Set up a collection — create a collection and import data into it'), and copy-pasteable examples ship in Python, JS/TS, Go, Java, and C# with realistic collection names and example responses.
- The single biggest discovery gap is a missing AGENTS.md — none exists in weaviate/weaviate, weaviate/docs, or the official client repos, even though Weaviate ships a dedicated weaviate/agent-skills repo for coding agents. An agent's coding-context tooling won't auto-discover repo conventions.
- Discovery is dragged down by absent llms-full.txt and .well-known/mcp.json (both 404), so agents relying on those exact conventions get nothing — though this is offset by an official Weaviate MCP server listed across PulseMCP, Smithery, and Glama.
- The OpenAPI spec is public and reachable (weaviate/weaviate openapi-specs/schema.json), so an agent can programmatically generate a typed REST client without scraping — a strong machine-readability signal.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Add an AGENTS.md to weaviate/weaviate (and the primary client repos) capturing build/test/convention context — the highest-weight discovery check currently failing (weight 3), and a quick win given the existing weaviate/agent-skills content to draw from.
- 02Publish a llms-full.txt at weaviate.io with the full markdown corpus of core API usage; the concise llms.txt exists but agents wanting comprehensive context hit a 404 (weight 1, but a common agent convention).
- 03Serve a valid .well-known/mcp.json advertising the official Weaviate MCP server's tools and endpoint, so agents can auto-discover it via the well-known path instead of only through third-party registries.
- 04Add an agent-facing error-recovery section: pair common REST/gRPC error codes with explicit remediation steps ('on 401, rotate the API key via…'; 'on schema conflict, check property casing') — the one comprehension check scoring only partial.
- 05Include the documentation pages in the base-domain sitemap (weaviate.io/sitemap.xml) or cross-reference the docs.weaviate.io sitemap, so an agent starting at the apex domain can enumerate the API docs rather than only marketing pages.
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.
Weaviate documents API-key and OIDC authentication with RBAC authorization, returns machine-parseable JSON errors via its REST/gRPC API, and documents both offset (limit/offset) and cursor (after) pagination. Idempotency is available through deterministic UUID generation and explicit object IDs rather than idempotency-key headers. Explicit API rate limits are not a documentation focus since Weaviate is primarily a self-hosted database; rate-limit handling appears mainly around third-party vectorizer providers.