Tavily is Agent-Ready to agents.
Discry independently scored how well an AI agent can discover and understand the Tavily 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 · 93/100Comprehension
55% of score · 100/100What we found
- An agent discovering Tavily lands on purpose-built infrastructure: a well-structured llms.txt index, a full llms-full.txt, a canonical agents.md setup guide, and every docs page available as clean Markdown by appending .md — this is reference-quality agent discovery, not an afterthought.
- The full OpenAPI 3.0.3 spec ('Tavily Search and Extract API') is publicly served through the Mintlify docs, and a valid /.well-known/mcp.json declares the remote MCP server with bearer + OAuth2 auth — two signals that ~98% of scanned APIs fail.
- Error recovery is best-in-class for agents: when a limit is hit, Tavily returns natural-language instructions telling the agent what happened and how to continue, alongside a documented 429 + retry-after contract — an agent can self-correct without a human in the loop.
- robots.txt explicitly allows every major AI crawler (GPTBot, ClaudeBot, anthropic-ai, Google-Extended, CCBot, PerplexityBot, ChatGPT-User), so agent-facing content is fully indexable.
- The one gap: no canonical AGENTS.md in the primary GitHub repo (tavily-mcp ships only a README). The hosted agents.md guide covers the same ground and more, but a coding agent inspecting the repo won't find the conventional file.
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 the primary GitHub repos (tavily-mcp, tavily-python, tavily-js) mirroring the hosted agents.md — this is the one discovery check not fully passing and would lift Discovery from 93 to ~100.
- 02Publish the OpenAPI spec at a stable, conventional path (e.g. https://docs.tavily.com/openapi.yaml or /openapi.json) in addition to embedding it in the Mintlify endpoint pages, so agents probing standard locations find it without scraping docs.
- 03Register the mcp.json / OpenAPI location within the docs sitemap and llms.txt so agents that start from the index can reach machine-readable specs in one hop.
- 04Document idempotency/retry-safety semantics explicitly (e.g., whether repeating a search on retry is billed and whether session_id de-duplicates), since agents building retry loops on top of the 429 guidance need to know the billing/side-effect contract.
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 documentation is exceptional: three auth paths (Bearer API key, OAuth 2.0 for MCP, and a keyless trial mode), machine-parseable JSON errors, and per-environment/per-endpoint rate-limit tables with an explicit retry-after header contract. No pagination or idempotency keys are documented — neither applies to Tavily's request/response search model (results are capped via max_results rather than paged).