Alpha Vantage is Not Agent-Ready to agents.
Discry independently scored how well an AI agent can discover and understand the Alpha Vantage 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 · 17/100Comprehension
55% of score · 53/100What we found
- An agent that already has the docs URL can use Alpha Vantage well — every endpoint ships copy-pasteable examples with a real ticker (IBM), real currency pairs (EUR/USD, BTC/EUR), a working 'demo' apikey, and code in Python, NodeJS, PHP and C#/.NET — but an agent discovering the API cold finds nothing: no robots.txt, llms.txt, sitemap.xml, or .well-known/mcp.json all return 404.
- Surprisingly strong agent intent despite zero discovery files: Alpha Vantage runs an OFFICIAL hosted MCP server (mcp.alphavantage.co) linked from the docs, ships an 'MCP & Other' code tab on every endpoint, and is listed across PulseMCP, Glama and other MCP registries — so an agent looking by MCP registry will find it even though file-based discovery fails completely.
- The entire reference is one ~780KB single-page document with no overview, no llms.txt, and no progressive disclosure — an agent must ingest a huge page to grasp core capabilities, blowing the token-efficiency budget even though a clean table of contents exists.
- Documentation is reference-only: it describes each endpoint in isolation with task-oriented summaries, but offers no multi-step workflow guides and no error-recovery guidance, so an agent learns what each call returns but not how to chain calls or self-correct after a failed request.
- Response payloads use Alpha Vantage's quirky numbered, space-containing JSON keys (e.g. 'Time Series (Daily)', '05. price') while request parameters are clean snake_case — an inconsistency an agent must handle when parsing results.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Add an llms.txt at the domain root summarizing the core API surface (auth, the query endpoint, key function families, the demo key) and linking to the documentation — the single highest-leverage discovery fix given the API is otherwise undiscoverable by file.
- 02Publish a robots.txt and sitemap.xml; both currently 404, which costs easy discovery points and leaves AI crawlers without explicit guidance to the documentation page.
- 03Publish an official OpenAPI/Swagger spec. Today only third-party reconstructions (Versori, Postman) exist; a first-party spec would let agents generate typed clients and validate calls instead of scraping a giant HTML page.
- 04Add a dedicated errors & rate-limits page documenting the JSON error/Note response shapes, the concrete free-tier limits (currently only on the pricing page), and explicit recovery steps (e.g., what to do on a rate-limit Note) so agents can self-correct.
- 05Add 2-3 multi-step workflow guides (e.g., 'find a ticker → pull daily series → compute an indicator') and an AGENTS.md in the official alpha_vantage_mcp repo to convert the strong endpoint reference into chainable, agent-oriented guidance.
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.
Auth is a single apikey query-string parameter (free key self-serve). The docs do not document the error response format, give no concrete rate-limit numbers (free-tier limits live on the pricing page, not the reference), and describe no pagination model — endpoints return full series gated by outputsize. No idempotency concept (read-only GET API).