Dwolla is Agent-Ready to agents.
Discry independently scored how well an AI agent can discover and understand the Dwolla 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 · 86/100Comprehension
55% of score · 100/100What we found
- An agent discovering Dwolla finds a near-complete signal set: a structured llms.txt, a 1MB llms-full.txt of clean markdown, a /.well-known/mcp.json pointing to a hosted MCP endpoint, a public OpenAPI spec repo, and a full sitemap — almost everything an agent needs to self-orient is served without scraping JS-rendered pages.
- Dwolla ships an official, registry-listed MCP server (github.com/Dwolla/dwolla-mcp, badged 'official' on Glama with a dedicated /docs/mcp-server page) exposing read-only account, transfer, and customer tools — an agent can act on Dwolla data through a first-party server rather than improvising HTTP calls.
- Comprehension is effectively flawless: endpoint descriptions are task-oriented ('Move funds between two bank accounts belonging to a single Verified Customer'), multi-step funds-flow guides use explicit Step sequences, and error docs give actionable recovery ('catch 401, request a fresh token, and retry') rather than bare status codes.
- The robots.txt uses the newer Content-Signal standard (ai-train=no, search=yes, ai-input=yes) and blocks no AI crawler — an agent reading docs at inference time is explicitly permitted, with only model-training use opted out.
- The only meaningful gap is the absence of an AGENTS.md in Dwolla's primary repos — a coding agent cloning an SDK gets no repo-level build/convention context, the single missing discovery signal keeping Discovery out of the A band.
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 SDK and MCP repos (dwolla-mcp, the language SDKs) with build/test/convention context — this is the one missing discovery signal and would lift Discovery from B to A.
- 02Verify and stabilize the /.well-known/mcp.json endpoint URL — it currently points at a 'dwolla.main-kill-isr.mintlify.me' preview-style host rather than a durable production domain, which risks breaking agent auto-discovery of the MCP server.
- 03Cross-list the OpenAPI spec directly from the developer portal (a stable /openapi.json or /openapi.yaml link) in addition to the dwolla-openapi GitHub repo, so agents resolving the spec from the docs domain don't have to infer the GitHub location.
- 04Consider trimming or splitting llms.txt (currently ~45KB) into a lean capability map plus the existing llms-full.txt, keeping the primary index comfortably under the agent-friendly size budget.
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 characteristics are thoroughly documented. Errors use a consistent JSON HAL envelope with machine-parseable error codes and an _embedded.errors array; rate limits (concurrency- and volume-based) include explicit 429 handling guidance; pagination is offset/limit based (default 25, max 200); and a dedicated Idempotency-Key page covers duplicate-request prevention. All claims are documentation-only and would need a live Discry Audit to verify.