Svelte is Good to agents.
Discry independently scored how well an AI agent can discover and understand the Svelte 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 · 62/100Comprehension
55% of score · 96/100What we found
- An agent asking an LLM about Svelte is exceptionally well served: svelte.dev pioneered the llms.txt convention and ships a full ladder of machine-readable docs — llms.txt (index), llms-small.txt, llms-medium.txt, and a 1.1MB llms-full.txt, plus per-package llms.txt for Svelte, SvelteKit, CLI, and Svelte AI. This is best-in-class token efficiency via progressive disclosure.
- A coding agent cloning the repo finds first-class guidance: AGENTS.md exists in both sveltejs/kit and sveltejs/ai, covering setup, build/lint/test commands, and coding style — a rising standard most frameworks still lack.
- Agent discovery through MCP is strong: there is an official Svelte MCP server plus multiple community servers listed across PulseMCP, Glama, and Smithery, so an agent can pull live, version-accurate Svelte docs and an autofixer.
- The scan's biggest deductions are structural, not quality: Svelte is a framework, so it has no OpenAPI spec (the single heaviest discovery check, weight 5), and it serves no sitemap.xml or .well-known/mcp.json — dragging Discovery to a C despite otherwise excellent signals.
- Comprehension is near-perfect: documentation is clean purpose-built markdown, examples use realistic values (real DB-backed load functions, not placeholders), capability boundaries are explicit (dedicated 'Caveats and limitations' sections), and multi-step onboarding is covered by an interactive tutorial and getting-started guides.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Add a sitemap.xml covering the docs, tutorial, and reference pages (currently 404). This is a trivial fix that lifts a weighted discovery check and helps crawlers enumerate documentation coverage.
- 02Publish a .well-known/mcp.json pointing to the official Svelte MCP server so agents can auto-discover the MCP endpoint from the domain rather than relying on third-party registries.
- 03Expand error-recovery guidance from descriptive to prescriptive: the compiler/runtime error references list what went wrong, but adding explicit 'to fix this, do X' steps for the most common errors would move errorRecoveryGuidance from partial to pass.
- 04Because Svelte is a framework (not a REST API), consider documenting this explicitly for agent tooling — an agent expecting an OpenAPI spec should be steered to the llms.txt ladder and MCP server instead.
- 05Cross-link AGENTS.md and the llms.txt files from the main docs navigation so both human contributors and agents discover them without a repo or well-known-path lookup.
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.
Svelte is a compiler/framework, not a hosted network API, so auth, rate limits, pagination, and idempotency do not apply. Error handling is well documented — SvelteKit exposes a generic error shape ({ "message": "Internal Error" }) with a handleError hook for custom reporting, plus dedicated compiler-error and runtime-error references and Svelte error boundaries.