Four checks an AI agent runs before a costly action — vet a dependency for CVEs and typosquats
before installing it, verify another x402 endpoint before paying it, coerce near-miss LLM
JSON to a schema, and turn a public URL into clean Markdown for RAG. Read-only, keyless, nothing stored.
Paid per call in USDC on Base via x402 or a prepaid credit key. Also a standalone MCP server.
Paid per call — $0.002 to $0.004.
Agents pay per call in USDC on Base via x402 (no account — an unpaid call returns a
signed 402 challenge to pay-and-retry). Humans can prepay by card and send an
X-FabTally-Key: fabt_… header instead. Bad input → 400, never charged.
Every check has a rate-limited free teaser (this page uses them).
HTTP (x402) — POST any check as JSON:
curl -XPOST https://guard.fabtally.com/v1/deps-check \
-d '{"ecosystem":"npm","name":"lodash","version":"4.17.20"}'
→ 402 with a signed x402 challenge; pay in USDC on Base and retry with the X-PAYMENT header.
Prepaid credit — add X-FabTally-Key: fabt_… to charge a card-funded balance instead.
MCP server — streamable-http: https://guard.fabtally.com/mcp · also stdio (npm: fabtally-guard-mcp).
4 tools (deps_check, x402_preflight, json_repair, url_read); paid tools relay the x402 challenge to pay-and-retry.
Discovery — /agent/overview · /openapi.json · /.well-known/x402