DecisionNode is Needs Work to agents.
Discry independently scored how well an AI agent can discover and understand the DecisionNode 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 · 52/100Comprehension
55% of score · 87/100What we found
- Domain mismatch: the queued domain decisionnode.com resolves to a BrandBucket 'For Sale' parking page, not the product. The real product lives at decisionnode.dev — an agent or human trying the .com first finds nothing usable. This scan assesses the actual product at decisionnode.dev.
- Strong discovery for a small tool: an agent integrating DecisionNode finds a genuinely well-structured llms.txt at decisionnode.dev — concise (~1.7KB), product-focused, with install commands and direct links to every docs page plus a single-file full-docs markdown. Also listed on the Glama MCP registry.
- Documentation is available as clean markdown (decisionnode-docs.md), so an agent gets copy-pasteable MCP config for Claude Code, Cursor, Windsurf and Antigravity plus realistic example decisions — without needing to render the JS site.
- The rendered /docs pages are a client-side SPA: an agent fetching decisionnode.dev/docs directly gets an empty shell. Only the llms.txt and the /decisionnode-docs.md markdown paths carry real content; llms-full.txt and .well-known/mcp.json both return the same empty shell.
- Error recovery is undocumented: an agent hitting a missing/invalid Gemini key, a quota error, or an embedding failure finds no troubleshooting guidance and would have to recover by trial and error.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Fix the domain — redirect decisionnode.com to decisionnode.dev (or publish a canonical pointer). The most-guessable domain currently resolves to a parking page, forfeiting all discovery for anyone who tries .com first. Highest-impact fix.
- 02Add an AGENTS.md to the GitHub repo (github.com/decisionnode/decisionnode) with setup and MCP-connection instructions for coding agents. Cheap win worth 3 discovery points and directly on-brand for a tool built for coding agents.
- 03Serve a real llms-full.txt (point it at the existing decisionnode-docs.md) and a .well-known/mcp.json manifest declaring the nine MCP tools and their schemas — both paths currently return the SPA shell.
- 04Add a troubleshooting / error-recovery section: what to do on a missing or invalid Gemini key, quota/rate-limit errors, and embedding failures (the existing `decide check` and `decide embed` commands can be documented as recovery steps).
- 05Document capability boundaries explicitly: Gemini free-tier rate limits, embedding/size limits, and what the tool intentionally does NOT do (not a rules engine, not injected into the system prompt).
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.
The only credential is a single Google Gemini API key (free tier) used for embeddings; the local MCP server runs over stdio with no auth. Error response formats, rate limits, pagination, and idempotency keys are not documented, though conflict detection at 75% similarity guards against duplicate or contradictory entries. All nine MCP tools have documented signatures (e.g. add_decision(scope, decision, rationale?, constraints?, global?, force?, project?)).