Running it
npm run serve # http://127.0.0.1:8787
npm run serve -- --port 9000 --rpc https://… # override| Tab | What it shows |
|---|---|
| Chain | Chain id, head block, gas price against the 20 Gwei floor |
| USDC index | Dual-emitter scan with the phantom-row and volume-inflation numbers, plus per-address reconciliation |
| Agents | Registry size, agent listing, owner and attestation detail |
| Lint | Paste Solidity or a deploy script, get findings with doc links |
| Docs | Ranked full-text search across every mirrored page, with the body inline |
A JSON API, not just a UI
Every tab is a thin client over routes that are just as usable from curl,
an editor extension, or CI.
curl -s localhost:8787/api/index/scan?blocks=150 | jq .stats
curl -s localhost:8787/api/docs/search?q=usdc+decimals | jq '.hits[0]'
curl -s -X POST localhost:8787/api/lint \
-H 'content-type: application/json' \
-d '{"lang":"solidity","source":"…"}'| Route | |
|---|---|
GET /api/chain | Chain id, head, gas price |
GET /api/index/scan?blocks=N | Canonical movements and dedup stats |
GET /api/index/reconcile?address=&blocks= | Indexed delta vs eth_getBalance |
GET /api/agents/count | Registry size via binary search |
GET /api/agents/list?from=&limit= | Owner and metadata URI per token id |
GET /api/agents/show?id= | One agent with its attestation summary |
GET /api/lint/rules | All 14 rules with severities and doc links |
POST /api/lint | {source, lang, filename} → findings |
GET /api/docs/search?q=&limit= | Ranked hits with marked excerpts |
GET /api/docs/page?site=&path= | One page, full markdown |
The documentation mirror
npm run mirror-docs # Arc → ./docs
npm run mirror-docs -- --site circle # Circle platform → ./docs-circle
The crawler starts from each site's llms.txt and follows internal links
breadth-first. Mintlify serves the markdown source of every page at
<path>.md, so the mirror is real prose rather than scraped HTML —
which is also why pages missing from llms.txt still land.
Search ranks title matches above summary matches above body frequency, and requires every term to appear. The boilerplate index banner Mintlify prepends to every page is stripped, or it would match everything.
Caching, because the RPC bites back
Arc's public RPC sits behind Cloudflare and answers bursts with 429. Every RPC-backed
route goes through a TTL cache that also collapses concurrent misses,
so a page load with four widgets makes one RPC pass rather than four. The agent count
— 41 eth_calls — is cached for five minutes, since the registry only
grows.
The server binds to 127.0.0.1 and has no authentication. It is a
developer tool, not a public service — do not expose it.
Tests
14 tests with no network: the documentation loader and ranking, the cache's TTL, coalescing, and failure behaviour, bigint serialisation, and request validation on every route.