Plaid is Good to agents.
Discry independently scored how well an AI agent can discover and understand the Plaid 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 · 69/100Comprehension
55% of score · 100/100What we found
- Plaid achieves a perfect comprehension score — docs serve native markdown (.html.md pages), have comprehensive error recovery guidance with per-error recommended actions, and task-oriented descriptions throughout
- Three-tier llms.txt system: /llms.txt (site index), /docs/llms.txt (API summary), /docs/llms-full.txt (complete docs in markdown) — the most sophisticated LLM-oriented content strategy scanned to date
- The error documentation is a gold standard for agent readiness: most common errors table with product applicability, linked to detailed pages with root causes and resolution steps
- OpenAPI spec publicly maintained at github.com/plaid/plaid-openapi, used to auto-generate official client libraries in 10+ languages
- robots.txt disallows /errors/* and /api/* paths which could impede some crawlers from indexing API reference content, though the llms.txt files compensate
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Add .well-known/mcp.json pointing to the official Plaid MCP server with tool declarations
- 02Add AGENTS.md to plaid/plaid-openapi and plaid/plaid-node repos with context for coding agents building financial integrations
- 03Consider removing the robots.txt disallow for /api/* to allow search engines to index the API reference pages
- 04Create a condensed llms-full.txt under 50KB focusing on core payment/auth flows for token-efficient agent consumption — current version at 6MB+ is too large for most context windows
- 05Add .well-known/mcp.json to plaid.com to enable automatic MCP server discovery
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.
Client ID + Secret API key authentication. Structured JSON error responses with error_type, error_code, error_message, and display_message fields. Rate limits documented with per-product details. Cursor-based pagination via /transactions/sync. Idempotency keys supported on mutation endpoints. Sandbox environment with test credentials for development.