Mercado Libre is Poor to agents.

Item: Mercado Libre API
Rating: 40
Author: Discry

Discry independently scored how well an AI agent can discover and understand the Mercado Libre API from what’s public — not whether it’s usable. Below: every signal we checked, what’s costing the score, and what to change.

Discry your API — free View the docs ↗

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 · 5/100

OpenAPI specA machine-readable OpenAPI/Swagger spec agents can parse.Fail

llms.txtAn llms.txt index that points agents to the docs that matter.Fail

llms.txt qualityThe llms.txt is focused, current, and well under the size budget.Fail

llms-full.txtA full-text bundle agents can load in one request.Fail

AGENTS.mdAn AGENTS.md that tells coding agents how to build on the API.Fail

.well-known/mcp.jsonA discoverable MCP manifest at a well-known path.Fail

MCP registryThe API is listed in a public MCP registry.Pass

robots.txt AI directivesrobots.txt allows (or explicitly guides) AI crawlers.Fail

SitemapA sitemap so agents can enumerate the docs surface.Fail

Comprehension

55% of score · 69/100

Task-oriented descriptionsEndpoints described by what they accomplish, not just their shape.Pass

Realistic examplesRunnable, real-world request/response examples.Partial

Multi-step workflowsDocs that chain calls into complete jobs an agent can follow.Pass

Error-recovery guidanceDocumented failure modes and how to recover from them.Pass

Answer-first formatThe answer leads; preamble does not bury it.Pass

Capability boundariesClear limits — what the API can and cannot do.Partial

Naming consistencyConsistent, predictable naming across endpoints.Pass

Heading hierarchyClean heading structure agents can navigate.Partial

Markdown docsDocs available as clean markdown, not JS-rendered HTML only.Fail

Token efficiencyDocs are concise enough to fit an agent context window.Partial

What we found

An agent hitting the canonical docs URL (developers.mercadolibre.com) gets nothing usable: it is a hard client-side SPA that returns only a 'Welcome / select your country' shell to any non-browser client, so robots.txt, llms.txt, sitemap.xml, and .well-known/mcp.json all resolve to that HTML rather than real files. Discovery scored 5/100 (F).
The actual documentation is genuinely good once reached — but only via country-mirror domains (e.g. developers.mercadolibre.com.ar / .com.gt) that render server-side. There an agent finds task-oriented endpoint descriptions, realistic curl examples with real item IDs (MLA599260060) and full JSON responses, and a complete end-to-end OAuth workflow.
The single reason an agent finds Mercado Libre programmatically is third-party effort: it runs an official hosted MCP server (mcp.mercadolibre.com/mcp) and is listed on PulseMCP, Glama, and Smithery — the sole passing discovery signal and the strongest agent-readiness marker in this scan.
Error recovery is a standout: the auth docs enumerate named error codes with concrete recovery steps (invalid_grant causes and fixes, a 4-step 'cannot connect' remedy, 'retry 429 after a few seconds', 403 cause guidance), letting an agent self-correct rather than fail blindly.
There is no public OpenAPI/Swagger spec, no AGENTS.md in the mercadolibre GitHub org, and no llms.txt — the highest-weighted discovery signals all fail, so an agent cannot machine-load the surface and must scrape scattered prose with no overview or progressive-disclosure summary.

What to change

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

01Serve real static files at the canonical domain: a valid robots.txt (explicitly allowing GPTBot/ClaudeBot), a sitemap.xml enumerating the /en_us/* doc pages, and an llms.txt indexing the core flows (auth, items, orders, shipments) with descriptions. This alone would lift discovery from ~5 to well above 40.
02Publish an official, downloadable OpenAPI 3.x spec for api.mercadolibre.com — the highest-impact single discovery fix (weight 5) and the artifact agents most want for typed-client generation.
03Make the primary developers.mercadolibre.com domain render documentation server-side (or offer content negotiation / a markdown mirror) instead of returning a country-selector shell to every non-browser agent.
04Add an AGENTS.md to the official mercadolibre-mcp-server repo and key SDK repos, describing tools, auth setup, and usage for coding agents — the repo currently ships only a bare README.
05Consolidate limits into one 'Rate limits & constraints' reference page (per-seller ~1500/min plus header names, multiget max 20, scan for >1000, limit max 100, regional availability) so agents discover boundaries by reading rather than by failing.

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.

OAuth 2.0 (Authorization Code grant, server-side)OAuth 2.0 PKCEBearer token✓ Error format documented✓ Rate limits documented✓ Pagination documented✕ Idempotency documented

Auth is OAuth 2.0 Authorization Code grant (server-side) with optional PKCE; access tokens expire in 6 hours and refresh tokens are one-time-use. Errors are documented as machine-parseable JSON with named codes (invalid_grant, invalid_client, invalid_scope, local_rate_limited, forbidden) and actionable recovery guidance. A rate limit is documented (~1500 requests/min per seller returning HTTP 429 local_rate_limited), though header names are not specified. Pagination uses offset/limit (paging object) plus a scan/scroll_id mode for >1000 records. Idempotency keys are not documented. All claims are documentation-only and unverified by live testing.

See your own Discry Score.

Drop your API docs URL. See what an agent sees — in 60 seconds, free.

Discry your API — free